ISS-10889: index every tracked file in a supported language; skip-list directories judge only untracked files

[wongk]

Implements ISS-10889.

What changes

A file git tracks, in a supported language, is now always indexed. The built-in directory-name skip list (node_modules, .venv, dist, build, data, results, …) applies only to files git does not track: untracked-but-not-ignored files, and every file in the non-git fallback walk.

Before, the skip list matched any path segment at any depth, tracked or not, so real source under a directory that happened to be called build, data or results was never indexed and never found by search, relations, code_query or code_changes.

Why

Tracking a file is the owner's statement that it belongs to the project. The name list only earns its place where git gives no signal: a repo that forgot to ignore .venv or node_modules must not flood the index with untracked dependency files.

Rules after this change

Rule	Tracked	Untracked / non-git
Supported language + extension gate	applies	applies
.lemoncrow/.ignore	applies	applies
Skipped-directory list	—	applies (list unchanged)
git-ignore	— (git never ignores a tracked file)	applies
1 MB file-size limit	applies	applies
Free-tier file cap	applies	applies

.lemoncrow/.ignore still binds tracked files: it is an explicit, user-written opt-out, and the CLI docs already describe it as the way to keep tracked data files out of the index.

Where

• inclusion.py: new git_tracked(root, candidates) (literal pathspecs so app/[id]/page.tsx is not a glob; submodules included; chunked at 1000 paths per call; fails open to "untracked", which applies the full rule list). exclusion_rule takes a required tracked set and skips the directory rung for tracked paths. indexable_paths (the per-edit re-index) asks git once per batch.
• graph.py: the whole-repo scan keeps which ls-files call listed each entry and applies the skip list only to the --others half.
• coverage.py: code_coverage_check names rules with the same tracked set, so it agrees with both entry points.
• engine.py: repo-scope symbol search no longer drops indexed hits by directory name (the index's admission rules are the only gate). Commit-history and external-scope hits keep the existing filter.
• docs/reference/cli.md: describes the tracked/untracked split.

Verification

• New tests: scan (tracked vs untracked under skip-list names), coverage + per-edit re-index agreement, .lemoncrow/.ignore still binding a tracked file, repo search returning a tracked build/ symbol, and git_tracked (literal brackets, submodules, at and past the chunk size, non-git fallback). Every new behavior test fails on main; the search test also fails with only the query-time filter reverted.
• Differential check with the real scan on two repos (old vs new iter_source_files, nothing dropped in either):
  • symphony-alpha: +8 files, exactly the tracked source under build/, data/ and results/ that was missing.
  • lemoncrow: +1 file, benchmarks/codebench/results/.../graft.json.
• Full fast suite passes. test_index_pool_does_not_fork_live_parent_state fails identically on main on this machine (spawn vs forkserver start method), so it is unrelated.