deps: move tree-sitter-language-pack to 1.x, with a clear error when a grammar cannot be downloaded - #156

Summary

Moves tree-sitter-language-pack from >=0.8.0,<0.9.0 to >=1.23.1,<2.0.0,
as agreed in #154. This is the first of two stacked PRs. The Lean 4 analyzer
follows on top of this branch.

Changes

• Pins. pyproject.toml has >=1.23.1,<2.0.0, and requirements.txt has
  ==1.23.1. tree-sitter==0.26.0 does not change. Pack 1.x declares
  tree-sitter>=0.23.
• Dart. The 1.x grammar parses Foo<T>(x) as a call whose
  argument_part holds type_arguments, not as a relational_expression.
  _process_chain now reads the type arguments from the call. Without this
  change, test_call_resolution_rules and test_riverpod_and_router_edges
  fail, because the Garage.open -> Item and carProvider -> Car edges are
  lost.
• Download at build time. docker/Dockerfile downloads the grammars the
  analyzers load (dart, go, javascript, typescript, vue, zig), so
  the image runs offline.
• Clear error with no network. Pack 1.x downloads each grammar on first
  use. Without network access, the first load raised a DownloadError that
  names a URL only. The per-file except blocks also caught it, so the
  language dropped out of the graph with a traceback for each file.
  • utils/grammars.py loads pack grammars through get_pack_parser. It
    changes the DownloadError into a GrammarUnavailableError that names the
    grammar, the cache folder and the t.download([...]) command.
  • analyze_code_files calls ensure_pack_grammars before the first file,
    so the run stops one time. The CLI shows the message through its usual
    Analysis failed: ... path.
  • PACK_GRAMMARS_BY_LANGUAGE lists the grammars that each language loads.
    A test checks that docker/Dockerfile downloads the same set.
• Docs. There is a CHANGELOG entry, and the "Adding a language" step in
  guides/development.md now uses get_pack_parser.

Example message, with an empty cache and no network:

The tree-sitter grammar 'go' is not in ~/.cache/tree-sitter-language-pack/v1.23.1/libs
and could not be downloaded (Download error: Failed to fetch manifest from
https://github.com/xberg-io/tree-sitter-language-pack/releases/download/v1.23.1/parsers.json:
io: Connection refused (os error 111)). tree-sitter-language-pack downloads each
grammar on first use, so the first analysis of these files needs network access.
To prepare a machine without network access, run this once on a machine with
access and copy the cache folder: python -c "import tree_sitter_language_pack as t;
t.download(['dart', 'go', 'javascript', 'typescript', 'vue', 'zig'])"

Tests

• tests/test_pack_grammars.py (new, 5 tests) covers the error message, that
  only the grammars of the present languages load, that the analysis stops
  before the first file, and that the Dockerfile list matches.
• The full suite gives 673 passed and 1 failed, on pack 1.23.1 and on 1.24.0.
  test_doc_write_guard.py::test_warning_reports_rejected_tool_calls also
  fails on main (a1b9f4a) in a fresh install, so it is not related to this
  change.
• ruff check and ruff format --check pass.
• docker build -f docker/Dockerfile . passes. In the image, get_parser
  loads all six grammars with --network none.

Out of scope: the image still needs network access at import time, because
codewiki/src/be/utils.py loads the tiktoken cl100k_base encoding when the
module is imported. This is already so on main. I can open a separate issue.

Note: tests/test_pack_grammars.py is added with git add -f, because
.gitignore has tests/*, like the other test files.

Refs #154

🤖 Generated with Claude Code