fix(openai): translate max_tokens to max_completion_tokens for reasoning models (#5847)

## What does this PR do?

OpenAI is deprecating `max_tokens` in favor of `max_completion_tokens`.
Reasoning models (o1, o3, o4-mini) and gpt-5+ already reject
`max_tokens` outright with:

```
Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead.
```

This PR translates `max_tokens` to `max_completion_tokens`
unconditionally for all OpenAI models, since all current models accept
`max_completion_tokens`. This avoids maintaining a hardcoded prefix list
that would need updating with every new model family.

A `DeprecationWarning` is emitted when the translation occurs, so users
are informed that their code should migrate to `max_completion_tokens`.
Code using `max_tokens` through OGX will continue to work, but the same
code run directly against OpenAI will fail for newer models.

The translation runs before the existing clamping logic and only applies
when `max_completion_tokens` is not already set by the caller.

Closes #5845

## Test Plan

Unit tests in `tests/unit/providers/inference/test_remote_openai.py`:

```bash
uv run pytest tests/unit/providers/inference/test_remote_openai.py -x --tb=short
```

```
tests/unit/providers/inference/test_remote_openai.py ...................... [100%]
22 passed in 0.29s
```

Tests cover:
- Translation applies to all models (gpt-4o-mini, gpt-4.1, o3, gpt-5,
dated variants)
- No translation when `max_completion_tokens` is already set by the
caller
- Original params object is not mutated
- Existing clamping logic still works after translation

Manually verified with a multi-model test script against a live OGX
server with `OPENAI_API_KEY`:
- Before fix: gpt-5, o1, o3, o3-mini, o4-mini all return 400
- After fix: all pass via `/v1/chat/completions` with deprecation
warning logged

---------

Signed-off-by: Adel Zaalouk <azaalouk@redhat.com>
Co-authored-by: Francisco Javier Arceo <arceofrancisco@gmail.com>
Co-authored-by: Matthew Farrellee <matt@cs.wisc.edu>

Author: 3 people

src/ogx/providers/remote/inference/openai/openai.py

@@ -4,6 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import warnings
 from collections.abc import AsyncIterator, Iterable
 
 from ogx.log import get_logger
@@ -113,6 +114,20 @@ async def openai_chat_completion(
         self,
         params: OpenAIChatCompletionRequestWithExtraBody,
     ) -> OpenAIChatCompletion | AsyncIterator[OpenAIChatCompletionChunk]:
+        # OpenAI is deprecating max_tokens in favor of max_completion_tokens.
+        # Reasoning models (o1/o3/o4) and gpt-5+ reject max_tokens outright.
+        # Translate unconditionally since all OpenAI models accept max_completion_tokens.
+        if params.max_tokens is not None and params.max_completion_tokens is None:
+            warnings.warn(
+                "max_tokens is deprecated by OpenAI and will be removed in a future release. "
+                "Use max_completion_tokens instead.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+            params = params.model_copy()
+            params.max_completion_tokens = params.max_tokens
+            params.max_tokens = None
+
         max_output_tokens = self._get_max_output_tokens(params.model)
         if max_output_tokens is not None:
             updated_params = params