checker: recognize table<...>/bare table as a table-shaped alternative

[Paritsingla7]

Summary

Found by actually running hyprvalidate check against two independent, real-world, distribution/project-scale Hyprland Lua configs — not synthetic testing.

• Garuda Linux's distribution-shipped settings (their Temeraire release, announced 2026-08-19, reworked around Lua config)
• sea-shell (a Hyprland desktop shell project, migrated to Lua in its 5.0.0 changelog)

Both are real, public, unauthenticated-fetchable repos — verified directly, not taken on faith.

The bug (hyprvalidate's own)

_has_table_alternative only checked for { or HL. in a type expression to decide "is this a table shape, not a scalar." It missed LuaLS's generic table<K, V> syntax and the bare table alternative entirely. HL.LayerRuleSpec.match is typed table<string, string|boolean> — so every real layer rule with a match table was a false positive, in both configs identically, since a match table is the only way to write a layer rule at all (7 false positives in Garuda's config, 4 in sea-shell's).

Fix: _has_table_alternative now also checks for "table" in the type expression. Verified: both real configs' layer-rule findings disappear completely; existing tests (including the HL.Gradient table-alternative case) still pass.

A second finding — upstream, not ours to fix

Both configs also use hl.bind(..., { mouse = true }), flagged as 'mouse' is not a field of HL.BindOptions. Traced directly into Hyprland's own source: mouse is a real, implemented flag (LuaKeybind.cpp, parsed identically to click/drag, which are in the schema) — it's just missing, along with auto_consuming, device_inclusive, and devices, from a hand-written field list in Hyprland's own meta/generateLuaStubs.py (lines ~658–677). That's the exact "hand-maintained table drifts from what's real" bug class this project exists to catch — just found one level further upstream. Not fixable here without hand-maintaining the same kind of table this project refuses to elsewhere. Noted in the CHANGELOG as a candidate for an upstream report — not filed yet, pending a decision on whether/how to do that.

Test plan

• pytest tests/ — 205 passed (3 pre-existing/environmental entry-point failures unrelated, deselected)
• Re-ran both real configs after the fix: the 11 layer-rule false positives are gone; only the genuine upstream mouse gap remains (2 findings, unchanged, correctly not something hyprvalidate can resolve)
• site/bundle.js regenerated (checker.py changed)