fix(parser): make python_parser.py keep_alive strategy tree-sitter 0.26-safe

[Wolfvin]

Mengapa

CI segfault (issue #235) di-mitigasi sementara dengan pin `tree-sitter<0.26` (PR #266). Root cause aktual: tree-sitter core 0.26.0 mengubah Node lifetime handling — binding tidak lagi buat Node objects hold strong reference ke Tree. `scripts/parsers/python_parser.py` (line ~135-165) sudah punya mitigasi `keep_alive` list + GC-disable, TAPI tidak cukup untuk 0.26: CI log (run 29244152939) menunjukkan native use-after-free SIGSEGV di `tree_sitter_python._binding` pada file shallow 250-baris (`test_file_above_old_threshold_is_parsed`), bukan yang deeply-nested.

Bukti: lokal (core 0.25.2) lolos; CI (core 0.26.0) crash. Delta = versi core.

Konteks

Selama cap `<0.26` aktif, CodeLens tidak bisa pakai tree-sitter 0.26+ (kehilangan perbaikan/grammar terbaru). Ini technical debt yang harus dilunasi.

Tujuan

Buat parsing (terutama `python_parser.py` iterative walk + keep_alive) aman terhadap perubahan Node lifetime di tree-sitter 0.26+, sehingga upper-bound `<0.26` di pyproject.toml + workflow bisa dilepas tanpa segfault.

Constraint

• Reproduksi WAJIB di Linux dengan tree-sitter core 0.26.0 (crash tidak muncul di Windows/0.25.2).
• Jangan cuma naikkan cap tanpa fix — harus buktikan test_large_file_parsing.py lolos di 0.26.
• Cek parser lain (js_backend, ts_backend, rust) yang pakai pola walk serupa — mungkin punya gap sama.

Definition of Done

• `test_large_file_parsing.py` lolos di Linux dengan tree-sitter core 0.26.x (no segfault)
• Cap `<0.26` dilepas di pyproject.toml + semua workflow yml
• Regression test yang gagal-tanpa-fix, lolos-dengan-fix

[Wolfvin]

🔄 Taking this issue — 2026-07-14T08:37:17Z

[Wolfvin]

🔄 Taking this issue — 2026-07-14T08:37:27Z

Identity: Executor·Pragmatist

Repo: CodeLens. Akan kerjakan langsung.

[Wolfvin]

PR siap review

PR: #286 — fix(parser): make keep_alive strategy tree-sitter 0.26-safe
Branch: fix/issue-267-python-parser-tree-sitter-0.26-safe
Summary: Root cause bukan keep_alive strategy saja — node.start_point / node.end_point access di tree-sitter 0.26 binding SIGSEGV non-deterministically (Point objects don't hold strong reference to Node/Tree). Fix di BaseParser.get_line (compute line dari start_byte via bisect on pre-computed line offsets) + new get_line_from_byte helper. Semua parsers (python, js_backend, ts_backend, tsx, rust) di-migrate off node.start_point / node.end_point. Additional fix di python_parser.py: pin ALL children to keep_alive at top of each iteration (0.26 binding tidak hold sibling refs). Cap <0.26 dilepas di pyproject.toml + 4 workflow yml.
Tests: test_large_file_parsing.py 11/11 passed (5/5 runs clean) · test_codelens.py + test_node_types.py + test_graph_model.py 141 passed · test_ts_backend_parser.py + test_js_backend_parser.py + test_rust_parser.py 62 passed — semua dengan tree-sitter 0.26.0. Runtime test di Linux (worker env). Pre-existing failure test_compact_format.py::TestSearchPagination bukan regresi (verified gagal juga di 0.25.2) — bagian dari #271.

Catatan pendekatan: issue #235 constraint bilang workflow YAML = BOS territory. Issue #267 DoD explicit bilang cap dilepas di workflow yml juga. Saya update workflow yml mekanis (hapus ,<0.26 + update comment) — BOS please review workflow yml changes carefully.