fix(docs): document JinjaX _attrs= for attrs passthrough (#78)

[fsecada01]

Summary

• docs/primitives.md documented <Cf:Button :attrs="{...}"> for JinjaX usage. JinjaX reserves the prop name attrs for its own extra-kwargs collector and unconditionally overwrites whatever a component's {#def} declares for it — that syntax compiles and runs but silently discards the dict under a real Catalog, with no error. Fixed the docs to show :_attrs="{...}" instead, JinjaX's own escape hatch for this exact collision, and explained why in a callout.
• Added tests/integration/test_jinja_attrs_passthrough.py, against a real jinjax.Catalog (mirroring test_jinja_autoescape.py's pattern). The existing unit-tier attrs tests never caught this bug: without a real Catalog, {#def} is a plain Jinja2 comment, so attrs never actually collides with anything there — only a real Catalog reproduces it, which is why these live at the integration tier.
• Investigating this surfaced a second, narrower gap: the RESERVED_ATTRS collision guard (PrimitiveConfigError) only reliably fires when the colliding key is not also one of the component's own declared prop names. For a key that's both (type/href for Button; name for the form controls), JinjaX's own arg-filtering routes it straight into that prop before render_attrs ever runs — the guard never fires, and the caller's real prop (if also passed) silently wins instead. This is a JinjaX-level limitation, not a cf-ui bug to patch around; documented explicitly in both the docs and the test module's docstring, and locked in with a test so a future JinjaX version change would be caught rather than assumed away.

Test plan

• pytest tests/integration/test_jinja_attrs_passthrough.py -v — 6/6 pass: bare kwargs passthrough, _attrs= dict passthrough, attrs= silently discarded (pins the bug), reserved-undeclared-name collision still raises, reserved-declared-name collision silently bypasses the guard (documented limitation), hostile value still escaped
• pytest tests/unit -q — 2826 passed, 13 skipped
• pytest tests/integration -q — 31 passed (25 existing + 6 new)
• mkdocs build --strict — clean
• ruff check / ruff format --check — clean

🤖 Generated with Claude Code

https://claude.ai/code/session_01Ur2c6d3peuerUGgRdJ6Fvs

[fsecada01]

/review findings

2 findings, both fixed in baa48eb:

1. [Fixed] docs/primitives.md's "Currently Button-only... (Primitives have no attribute-passthrough mechanism (Button, FormField, Select, Textarea, etc.) #71, Spike: design an attribute-passthrough mechanism for primitives #72)" line was stale — both issues are closed and Roll out attrs passthrough to Select, Textarea, FormField #76/feat(select,textarea,form-field): roll out attrs passthrough (#76) #77 already rolled attrs passthrough out to Select/Textarea/FormField, which directly contradicted this same PR's own new text ("name for the form controls") two sentences earlier. Corrected to name what actually ships today.

2. [Fixed] The declared-prop-name collision-bypass gap (a key that's both RESERVED_ATTRS-listed and a component's own declared prop silently bypasses the PrimitiveConfigError guard) was only pinned by a Button test, but both the docs and the test module's docstring claimed it for "the form controls" generally. Added test_the_same_declared_prop_name_gap_holds_for_a_form_control (FormField's name= collision) so a future JinjaX change affecting a different prop signature would actually be caught.

All gates re-verified: pytest tests/unit (2826 passed, 13 skipped), pytest tests/integration (32 passed — 25 existing + 7 attrs tests), mkdocs build --strict clean, ruff check/ruff format --check clean.