spytial-core 3.2.3: draw endpoints accept unary groups; textStyle large

[sidprasad]

Bumps the pinned spytial-core from 3.2.1 to 3.2.3 and corrects the inferredEdge docs that 3.2.2 makes wrong. Two upstream releases landed while this was open, so the PR now tracks both.

What changed

./update-spytial-core.sh handled the mechanical part of each bump: the CDN pin in core_assets.py, the vendor README pin, and the IJsonDataInstance test docstring. Both releases changed only the browser bundles, so the re-vendored tier-2 evaluator came back byte-identical each time and _vendor/spytial-core-evaluator.js is not in this diff. Patch bump to 1.8.1 covers the PR as a whole.

3.2.2 — core#501, and why the docstrings changed

#501 extends the inferredEdge draw feature we shipped in 3.2.1:

• a draw endpoint may now name a unary group constraint — it attaches to that single group directly, and the end's atom plays no part. Previously this parsed fine but silently dropped every edge at layout time.
• an edge whose ends land on the same group is kept and drawn as a self-loop on the hull, rather than dropped.
• a name meaning both a keyed group and a single group is now a layout-time error (with no key in play, there's no way to pick).

Our docstring stated the opposite as a rule — that a draw endpoint's group "must be keyed", and that a unary group's edge "is dropped without drawing". That's the one kind of stale doc worth catching in a version bump: not vague, but a confident instruction pointing users away from something that now works. Both the inferredEdge decorator doc and the InferredEdge class doc are rewritten to cover keyed vs. unary endpoints, the self-loop, and the new ambiguity error.

_validate_draw needs no change — it checks only the '<end> -> <end>' shape, which #501 leaves alone. The existing shape tests in test_decorator_docs.py still cover it.

3.2.3 — core#502

Makes textStyle: { size: large } render perceptibly larger than the node label (tier constant 16px → 20px; edge labels no longer have the authored tier overwritten by the zoom handler's hardcoded 12px base each layout tick).

No doc change needed on this side: our TextStyle docs in annotations.py and docs/operations.md only enumerate 'small' | 'normal' | 'large', never pixel values or relative sizes, so nothing here described the behavior #502 corrected.

Reviewer notes

• No behavior change in this package; the semantics live in core. The only judgment call is the draw doc rewrite — worth a read against #501's description.
• Verified: 444 tests pass. I also ran the four opt-in real-browser e2e tests (SPYTIAL_BROWSER_TESTS=1 pytest test/test_edit_browser.py), the ones that actually load the pinned bundle from the CDN — all pass against 3.2.3, and the three jsdelivr asset URLs return 200.
• One flake worth knowing about: the first browser run right after the 3.2.3 bump failed (test_done_round_trips), then passed on the next three full-file runs. The failing run took 39s against ~6s for the clean ones — it raced a cold jsdelivr cache for the newly published 3.5MB bundle, since the reachability check at test_edit_browser.py:49 reads only 1 byte and doesn't warm it. That's a pre-existing first-run-after-bump fragility rather than anything about 3.2.3, but I'm flagging it rather than burying it.

🤖 Generated with Claude Code