fix: choose a file's grammar by its [formats] or --ext language

[iainlane]

Vale decides how to read a file from its extension. In a source file, it lints only the comments. To find the comments, it parses the file with a tree-sitter grammar for the file's language. Clojure and PowerShell have no grammar, so Vale finds their comments with regular-expression comment patterns instead. Files with unknown extensions are linted as prose.

A user can tell Vale to read a file as another language, either with the [formats] section of .vale.ini or with the --ext flag. Users do this for languages that Vale does not support, such as Nix, by choosing a supported language with the same comment syntax. With nix = pl in [formats], Vale should lint the comments in default.nix and skip the Nix code.

Today, this only works when Vale finds the target language's comments with patterns. When the target language has a grammar, Vale skips every comment in the file. This limited coverage is a regression: before 3.5.0, every code language had comment patterns, and reading a file as any of them worked. Vale removed each language's patterns when the language moved to tree-sitter: Python, Rust, Ruby, Go, JavaScript, TypeScript, C++ and Protocol Buffers in 3.5.0, and the rest apart from Clojure and PowerShell in 3.18.0 (e82432d, "extract Lua, PHP, Haskell, R, and Perl comments with tree-sitter").

lintCode looks up the grammar by the file's real extension, not by the target language's extension. .nix has no grammar, so lintCode falls back to lintCodeOld, which looks up comment patterns by the target language's extension. Perl has no patterns, so lintCodeOld returns without linting any comments. The same lookup means that the target language has no effect when the real extension has a grammar: under --ext=.c, Vale still parses a .ts file as TypeScript. For markup, and for the comment patterns, the target language decides how Vale reads the file.

The lookup by real extension dates from 172714d ("deprecate regex-based comment parsing"). At that time, Go, JavaScript, TypeScript and several other languages shared the normalized extension .c, so a lookup by the normalized extension would have parsed them all as C. Each of these languages now has its own normalized extension, so the two lookups give the same grammar unless the user has chosen a target language.

This PR makes lintCode look up the grammar by the target language's extension. lintDataComments, which checks for a grammar before it calls lintCode to lint a data file's comments, does the same. Files without a [formats] entry or --ext are parsed as before. A user who maps one language with a grammar to another, such as ts = js, now gets the target language's grammar.

The new e2e cases lint/perl-format-mapping, lint/perl-ext-flag, lint/ext-flag-overrides-grammar and views/data-comments-format-mapping fail on v3 and pass with this change.

I wrote this change with Claude Code.

[github-actions]

Before this is reviewed, please read the AI-Assisted Contributions section of the contributing guide and make sure this change meets it: you can explain every line, the description is your own and brief, it fixes a bug you hit or a feature agreed in an issue, it does not touch undocumented options, and any substantial tool use is noted.

[github-actions]

The license agreement is signed.
_{Posted by the CLA Assistant Lite bot.}

[iainlane]

I have read the CLA and I agree to it