feat(lint): Upgrade ty (#543)

* squash
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

* fix(lint): suppress new ty 0.0.56 error categories in CI

The ty upgrade from 0.0.17 to 0.0.56 catches additional pre-existing
type violations. Add these new error categories to the CI ignore list
with counts, matching the existing pattern for incremental cleanup.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

* fix(lint): add --exit-zero-on-warning for ty 0.0.56

ty 0.0.56 exits non-zero on warnings (unsupported-base,
unused-type-ignore-comment, redundant-cast). Add
--exit-zero-on-warning so only errors fail CI.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

* fix(lint): resolve all new ty 0.0.56 errors at source

Instead of adding blanket suppressions for new error categories,
fix each violation individually:

- Fix with_logging decorator in anonymizer/data-designer to preserve
  class types through TypeVar (invalid-type-form)
- Fix StudioService.dependencies to use ClassVar (invalid-attribute-override)
- Add inline ty: ignore comments for false positives and intentional
  test misuse (invalid-paramspec, call-non-callable, unknown-argument,
  missing-argument, empty-body, too-many-positional-arguments,
  invalid-yield, unresolved-import, invalid-type-form)

No new blanket suppressions added to lint-python-types.sh.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

* fix(lint): consolidate duplicate type ignore comments

Replace `# type: ignore[rule]  # ty: ignore[rule]` with blanket
`# type: ignore` on lines that need both mypy and ty suppression.
Blanket `# type: ignore` is respected by both checkers, eliminating
the need for separate directives.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

* Revert "fix(lint): consolidate duplicate type ignore comments"

This reverts commit cd7b6f6.

Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

* fix(lint): use ty: ignore instead of type: ignore

This repo uses ty, not mypy. Replace redundant dual comments
(# type: ignore[rule]  # ty: ignore[rule]) with just # ty: ignore[rule]
on the 14 lines we touched for the ty upgrade.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

* chore: sync vendored SDK and CLI after ty ignore cleanup

Run make vendor and make update-cli to sync the evaluator SDK
vendored files with the ty: ignore comment changes.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>

---------

Signed-off-by: Matthew Grossman <mgrossman@nvidia.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: matthewgrossman

packages/nemo_evaluator_sdk/src/nemo_evaluator_sdk/resilience/api.py

@@ -69,7 +69,7 @@ async def run_with_resilience(
     endpoint_key: str,
     operation: Callable[_P, Awaitable[_T]],
     *args: _P.args,
-    max_attempts: int,
+    max_attempts: int,  # ty: ignore[invalid-paramspec]
     deadline_at: float | None = None,
     **kwargs: _P.kwargs,
 ) -> _T:

packages/nemo_evaluator_sdk/src/nemo_evaluator_sdk/resilience/scheduler.py

@@ -105,7 +105,7 @@ async def run_with_resilience(
         endpoint_key: str,
         operation: Callable[_P, Awaitable[_T]],
         *args: _P.args,
-        max_attempts: int,
+        max_attempts: int,  # ty: ignore[invalid-paramspec]
         deadline_at: float | None,
         **kwargs: _P.kwargs,
     ) -> _T:
@@ -202,7 +202,7 @@ async def _run_once(
         endpoint_key: str,
         operation: Callable[_P, Awaitable[_T]],
         *args: _P.args,
-        attempt: int,
+        attempt: int,  # ty: ignore[invalid-paramspec]
         **kwargs: _P.kwargs,
     ) -> _T:
         controller = await self._get_controller(endpoint_key)

packages/nemo_evaluator_sdk/tests/metrics/test_llm_judge.py

@@ -522,7 +522,7 @@ async def test_rubric_compute_scores_emits_numeric_score_and_label(self, mocker:
 
     def test_unique_scores_allows_empty_scores_for_constructed_model(self):
         metric = LLMJudgeMetric.model_construct(model=_make_model(), scores=[], _fields_set={"model", "scores"})
-        assert metric.unique_scores() is metric  # type: ignore[call-non-callable]  # Pydantic model_validator descriptor
+        assert metric.unique_scores() is metric  # ty: ignore[call-non-callable]  # Pydantic model_validator descriptor
 
     def test_string_prompt_template_bypasses_reserved_key_validation(self):
         metric = LLMJudgeMetric(

packages/nemo_evaluator_sdk/tests/test_agent_inference.py

@@ -106,7 +106,7 @@ def test_api_key_none_when_no_secret(self):
 
     def test_extra_fields_forbidden(self):
         with pytest.raises(ValueError):
-            Agent(url="http://agent.test", name="test", format=AgentFormat.NEMO_AGENT_TOOLKIT, extra_field="bad")  # type: ignore[call-arg]
+            Agent(url="http://agent.test", name="test", format=AgentFormat.NEMO_AGENT_TOOLKIT, extra_field="bad")  # ty: ignore[unknown-argument]
 
     def test_generic_with_trajectory_path(self):
         agent = Agent(

packages/nemo_platform_plugin/src/nemo_platform_plugin/_spec_flags.py

@@ -326,7 +326,7 @@ def _extract_default(info: FieldInfo) -> Any:
         return info.default
     if info.default_factory is not None:
         try:
-            return info.default_factory()  # type: ignore[call-arg]
+            return info.default_factory()  # ty: ignore[missing-argument]
         except TypeError:
             # Pydantic supports validated_data-aware factories; we can't
             # safely call those at CLI-build time. Treat as required so
@@ -451,7 +451,7 @@ def _optional_of(tp: object) -> Any:
     runtime value. Delegating to ``__getitem__`` here keeps the annotation
     correct without tripping static type checkers.
     """
-    return Optional[tp]  # type: ignore[valid-type]
+    return Optional[tp]  # ty: ignore[invalid-type-form]
 
 
 def make_field_param(

packages/nemo_platform_plugin/src/nemo_platform_plugin/function.py

@@ -180,7 +180,7 @@ class NemoFunction(_NamedPlugin, Generic[SpecT]):
     # ``spec_schema`` to ``type[GreetSpec]`` for type-checkers and IDEs.
     # PEP 526 forbids type variables inside ``ClassVar``, so the suppression
     # here is the standard escape hatch for the otherwise-correct shape.
-    spec_schema: ClassVar[type[SpecT]]  # type: ignore[misc]
+    spec_schema: ClassVar[type[SpecT]]  # ty: ignore[invalid-type-form]
 
     # ------------------------------------------------------------------ #
     # Endpoint override                                                  #

packages/nemo_platform_plugin/tests/test_dispatcher.py

@@ -170,7 +170,7 @@ def test_returns_1_for_none_return(self, monkeypatch, tmp_path: Path) -> None:
         class _Job(NemoJob):
             name = "forgot-return"
 
-            def run(self, config: dict) -> dict:  # type: ignore[return]
+            def run(self, config: dict) -> dict:  # ty: ignore[empty-body]
                 # Intentionally no return — most plausible failure mode.
                 pass
 

packages/nemo_platform_plugin/tests/test_function_context.py

@@ -30,7 +30,7 @@ def test_with_request_id() -> None:
 def test_kw_only_construction() -> None:
     """Workspace must be passed by keyword — the dataclass is ``kw_only``."""
     with pytest.raises(TypeError):
-        FunctionContext("default")  # type: ignore[misc] — intentional misuse
+        FunctionContext("default")  # ty: ignore[missing-argument,too-many-positional-arguments]  — intentional misuse
 
 
 def test_field_set_is_minimal() -> None:

packages/nemo_platform_plugin/tests/test_inference_middleware.py

@@ -394,7 +394,7 @@ def test_config_id_reference(self):
 
     def test_config_type_required(self):
         with pytest.raises(ValidationError):
-            MiddlewareCall(name="nemo-switchyard")  # type: ignore[call-arg]
+            MiddlewareCall(name="nemo-switchyard")  # ty: ignore[missing-argument]
 
 
 # ---------------------------------------------------------------------------

packages/nemo_platform_plugin/tests/test_job_context.py

@@ -58,4 +58,4 @@ def test_results_is_required(self, tmp_path: Path) -> None:
         # Omitting ``results`` raises at construction; the dataclass
         # enforces it (no silent ``None`` handed to the job).
         with pytest.raises(TypeError, match="results"):
-            JobContext(workspace="ws", storage=_make_storage(tmp_path))  # type: ignore[call-arg]
+            JobContext(workspace="ws", storage=_make_storage(tmp_path))  # ty: ignore[missing-argument]