feat: infer language for untagged code blocks

[denolfe]

Untagged fenced code blocks get syntax highlighting when their content clearly matches a bundled grammar. Blocks that match no rule stay plain.

Key Changes

• Rule-based language detection
  • A new detectLang module checks block content against high-precision rules. The first match wins, and no match returns undefined.
  • Rules cover shebang interpreters, a $ shell prompt, valid JSON, HTML, Go, Python, Rust, and TOML.
  • HTML matches a doctype or a short list of lowercase tags. Blocks containing ={ (a JSX expression attribute) are excluded.
• Inferred language drives highlighting
  • Code nodes carry the guess as inferredLang. It is set only when the fence has no info string (the text after the opening backticks).
  • The guess picks the grammar. The block title shows only what the author wrote.
  • Render mode (--render) preloads the inferred grammar too.
• Test fixture
  • test/lang-detection.md lists detected, plain, and tagged cases. Each heading states the expected result.

The following prior art was investigated, but rejected for various reasons:

• flourite: regex-based detector, evaluated and rejected. It labeled file trees, logs, and prose as Lua, and it has no bash, JSON, or TOML support.
• highlight.js: highlightAuto evaluated and rejected. Short snippets came back as nix, applescript, and mipsasm.
• GitHub Linguist shebang parser: source of the shebang interpreter rules.

[github-actions]

Startup benchmark (--render test/exhaustive.md, linux-x64)

build	mean	ratio	verdict
baseline (main@714aff7)	294.0ms ± 5.7ms	—
PR	300.6ms ± 13.5ms	1.02×	✅ ok

Thresholds: warn ≥ 1.1×, fail ≥ 1.25×. Baseline built from main.

First-frame (large.md, runs=10)

metric	baseline (main@714aff7)	PR	ratio	verdict
first-frame	213.0ms	226.1ms	1.06×	✅ ok

Thresholds: warn ≥ 1.1×, fail ≥ 1.25×. Baseline built from main.

Scroll (large.md)

metric	baseline (main@714aff7)	PR	ratio	verdict
full_mount	10338.0ms	10594.0ms	1.02×	✅ ok
step_p50	27.0ms	27.4ms	1.01×	✅ ok
step_search_p50	28.4ms	28.5ms	1.01×	✅ ok
stepMatch_p50	21.3ms	21.5ms	1.01×	✅ ok
commit	129.0ms	130.8ms	1.01×	✅ ok

Thresholds: warn ≥ 1.1×, fail ≥ 1.25×. Baseline built from main.