fix(extract): converge Function/Method is_test with the tests/ path filter

[Yyunozor]

Closes #1294.

Summary

• Function/Method.is_test only ever became true via the Rust #[test] attribute check; the per-file test-file flag (cbm_is_test_file) was already computed and stored on the Module node, but never propagated to the definitions inside it. A C function in tests/helpers/fixtures.c stayed is_test=false and leaked past store.c's is_test != 1 filters, even though trace_path's own independent path check already treated the file as a test.
• extract_func_def and push_method_def now OR ctx->result->is_test_file into def.is_test, additive to the existing Rust attribute check (#[test] functions embedded in an otherwise regular .rs file are unaffected).
• cbm_is_test_file (extraction-time) gained the same tests//test//spec//__tests__/ directory-prefix convention the project's own cbm_is_test_path (pass_tests.c) already uses, so a file under tests/ with no test_/_test naming is caught too — applying an existing convention where it was missing, not inventing a new one.
• Secondary bug: trace_call_path's own include_tests filter (mcp.c's separate, query-time is_test_file()) matched a nested .../tests/... path but not a project-root-relative one, so tests/repro/rpd_index_and_fingerprint leaked into results under the default include_tests=false. Added the same missing directory-prefix check there.

Verification

• make -f Makefile.cbm test -j12: parent 6786 passed / 1 failed / 4 skipped; fix branch 6789 passed / 1 failed / 4 skipped. Delta is exactly the 3 new tests; the 1 pre-existing failure (test_cli.c:8526, Copilot CLI/VS Code detection) and 4 skips (Windows-only) are identical on both sides.
• New tests: extract_c_test_dir_marks_is_test_issue1294, extract_python_method_test_dir_marks_is_test_issue1294 (positive/negative for both extraction call sites) and tool_trace_totals_respect_test_filter_tests_root_subtree_issue1294 (end-to-end trace_call_path include_tests toggle).
• clang-format/clang-tidy/cppcheck are not available in my local environment (clang-format errors on a libz3 version mismatch in my Homebrew LLVM), so CI is the authority for lint, as in fix(extract-rust): mark #[test]/#[tokio::test] fns is_test so they are filtered #857. The build itself uses -Wall -Wextra -Werror and compiled clean on both branches.

Known limitations

• Path matching (here and in the pre-existing cbm_is_test_path) is forward-slash-only and case-sensitive; unchanged, to keep this diff minimal.
• store.c's cbm_is_test_file_path() (a third, broader strstr(fp,"test") implementation feeding search_graph/dead-code) is untouched — outside this issue's reported scope.
• A natural follow-up: unify the four divergent is-test-path implementations (pass_tests.c, helpers.c, mcp.c, store.c) behind one shared function instead of a third copy of the same directory list — left for a separate PR.

[DeusData]

Thank you for the contribution and for tracing the inconsistent test classification through extraction and MCP filtering. This is now triaged as a high-priority parsing-quality bug for 0.9.1-rc. Our community PR queue is currently quite full, so it may take a little time before we can complete the review and, if approved, merge it. We are doing our best to support community contributions and will return with code-grounded feedback as capacity opens.