fix: preserve Kotlin annotations + resolve C# namespace imports (#295, #310)

[azizur100389]

Summary

Two parser correctness fixes that restore metadata and resolution paths lost by v2.3.2.

Closes #295, closes #310.

---

Fix 1 — Kotlin (and Java/C#/Python) annotation preservation (#295)

_extract_functions already walked tree-sitter modifiers > annotation children to collect decorators, but used the list only for test-kind detection and then discarded it. Kotlin @Composable, @HiltViewModel, @Inject; Java @Entity, @Autowired; C# [HttpGet]; Python decorators — all were missing from the persisted node.

_extract_classes did not extract class-level annotations at all, so @HiltViewModel class MyViewModel, @AndroidEntryPoint, etc. lost their annotation metadata entirely.

Fix:

• New module-level _extract_annotation_list(node) helper that centralises the Java/Kotlin/C# modifiers walk and the Python decorated_definition walk.
• _extract_functions now persists annotations to node.modifiers (comma-separated string) AND node.extra["decorators"] (list) on every Function / Test node.
• _extract_classes calls the same helper to preserve class-level annotations.

Consumers can now write queries like "show me all @Composable functions" or "find @HiltViewModel classes" — both currently impossible on Kotlin codebases.

Fix 2 — C# namespace-based importers_of resolution (#310)

C# using ACME.Core; directives produce an IMPORTS_FROM edge whose target is the raw namespace string ("ACME.Core"), not a file path. importers_of looked edges up by file path and never matched a namespace, so it returned [] for every .cs file — breaking get_impact_radius, detect_changes, and dead-code analysis on C# codebases.

Fix:

• New _extract_csharp_namespaces(root_node) helper handles both tree-sitter shapes:
  • block form: namespace_declaration
  • C# 10+ file-scoped: file_scoped_namespace_declaration
• parse_bytes tags each C# File node with the namespaces it declares in extra["csharp_namespaces"].
• tools/query.py::query_graph adds a namespace fallback to importers_of: when the target is a .cs file, look up its declared namespaces and also search edges_by_target for each namespace string. Deduped against the primary lookup so duplicates never appear.

This mirrors the existing bare-name fallback already used by callers_of and inheritors_of (see #87). No parser restructuring needed — just an additional lookup path in the query layer.

---

Tests added (10 new tests in tests/test_multilang.py)

TestKotlinAnnotations — 5 tests:

Test	Purpose
test_hilt_viewmodel_annotation_on_class	@HiltViewModel persisted on Class node
test_composable_annotation_on_function	@Composable persisted on Function node
test_multiple_annotations_on_function	multiple @Inject + @field:JvmField both preserved
test_unannotated_function_has_none_modifiers	regression guard: no leak of empty string / empty list
test_test_annotation_still_triggers_test_kind	guard: pre-existing @Test → Test kind promotion still works

TestCSharpNamespaceResolution — 5 tests:

Test	Purpose
test_file_scoped_namespace_tagged_on_file_node	C# 10+ namespace Foo; form
test_block_namespace_tagged_on_file_node	classic namespace Foo { ... } form
test_multiple_namespaces_in_one_file	both namespaces kept
test_non_csharp_file_has_no_namespace_tag	scope guard
test_importers_of_resolves_namespace_to_file	end-to-end: write Core.cs + App.cs + Unrelated.cs, query importers_of Core.cs, assert App.cs is found and Unrelated.cs is not

Test results

Stage	Result
New targeted tests	10/10 passed
tests/test_multilang.py full	203 passed, 38 pre-existing failures (verified identical on unchanged main)
Full suite	1016 passed, 73 pre-existing failures (verified identical set on main via comm -23 diff — zero regressions)
ruff check on all 3 changed files	clean
mypy on parser.py + query.py	clean

Zero regressions. All fixes follow patterns already used in the codebase (inheritors_of bare-name fallback, File node extra tagging mirroring the Swift swift_kind pattern).

[azizur100389]

Heads up on CI: the 3 failing checks (lint, type-check, test) are pre-existing baseline failures on upstream/main, not regressions from this PR. Verified by running each check on my branch and on a clean checkout of b0f8527 (the merge base) and getting identical failure sets.

Check	Failure source	Mine?
lint	9× N999 Invalid module name for accidentally-committed macOS Finder duplicates (memory 2.py, analysis 2.py, enrich 2.py, exports 2.py/3.py, graph_diff 2.py, jedi_resolver 2.py, token_benchmark 2.py, memory 3.py) + 1× I001 import sort in main.py:9 + 1× F401 unused import in tools/review.py:9	No
type-check	main.py:945: "FastMCP[Any]" has no attribute "_tool_manager"	No
test (3.10/.11/.12/.13)	61 failures, including 6 errors in test_main.py::TestApplyToolFilter from the same _tool_manager issue, plus the same set that fails on unmodified main	No

All three are addressed by @gzenz's existing #342 "chore: fix CI lint and type errors on main".

For the actual changes in this PR:

• 10/10 new regression tests in TestKotlinAnnotations and TestCSharpNamespaceResolution pass locally on Python 3.10
• Full tests/test_multilang.py shows 38 failures = same set that fails on unmodified main (zero regressions, verified via comm -23 diff)
• ruff check clean on all 3 files I touched (parser.py, tools/query.py, tests/test_multilang.py)
• mypy clean on parser.py and tools/query.py

Happy to rebase once #342 lands.