fix: don't crash when an aerial query file fails to parse

[Booyaka101]

Closes #506.

vim.treesitter.query.get(lang, \"aerial\") raises whenever the query file references node types absent from the installed grammar. The current reported case: SQL files crash aerial with

Query error at 37:2. Invalid node type \"create_policy\":
(create_policy
 ^

because queries/sql/aerial.scm uses create_policy, which was added to tree-sitter-sql in v0.3.10 (2026-01) and is missing from older parsers (very common when using nvim-treesitter master branch on Neovim < 0.12 — see DerekStride/tree-sitter-sql#347).

The fix

Wrap query.get in pcall and return nil on parse failure inside get_query. Both existing callers already handle nil (init.lua:22 in is_supported; init.lua:184 in get_lang_and_query), so aerial just treats the language as treesitter-unsupported for that one user instead of crashing on every buffer enter. Emit a single user-facing warning; cache the failure so the warning doesn't spam on every buffer.

This is intentionally a defensive layer only — I'm not removing create_policy from queries/sql/aerial.scm since that was added on purpose in #505 and might be used by people on newer grammars. If you'd like to additionally restructure that file (e.g. split into a stable core + extras, or drop create_policy to keep partial outline working on old grammars), happy to follow up in a separate PR.

How this was tested

End-to-end reproduction with a synthetic queries dir that references a non-existent node type for an installed grammar (lua):

	Behavior
Baseline	`FAIL: get_query raised: ...Query error at 5:2. Invalid node type "__deliberately_missing_node_simulating_create_policy"`
With fix	`aerial: failed to load treesitter query for lua, treating as unsupported...` warning logged, `get_query` returns `nil`

3 regression tests in `tests/helpers_spec.lua`:

• returns nil instead of crashing when the query fails to parse
• caches the parse failure so the warning only fires once (across repeated calls)
• still returns the query when parsing succeeds

All 3 pass with the fix; 2/3 fail on baseline (the success-case test passes either way).

[stevearc]

This has the look of minimally edited LLM output. Just from skimming the changes that I initially want to see are:

• remove the unnecessary, overly verbose comments
• instead of logging these errors, make them surface in the :AerialInfo diagnostics
• ensure that all tests are valuable and we're not just testing trivial logic

[Booyaka101]

Thanks, pushed an update — errors now flow through is_supported so they show in :AerialInfo. Also trimmed the comments and dropped the trivial test.

[stevearc]

LGTM, thanks for the PR!