claude-code-chat-browser: Extract JSONL parser monolith into focused modules

[clean6378-max-it]

Problem

utils/jsonl_parser.py (~761 lines) combines three distinct responsibilities:

1. JSONL line-level parsing — parse_session() read loop, message processors, metadata assembly
2. Tool-result dispatch — 15-predicate ordered table _TOOL_RESULT_DISPATCH (lines ~570–586) where first match wins; ordering is load-bearing and documented only with inline comments
3. Session peek optimization — quick_session_info() for fast title/timestamp without full parse

This coupling makes dispatch ordering harder to reason about, increases merge-conflict risk, and raises the cost of adding new tool-result shapes. Tuesday's real-session fixtures and dispatch-order regression tests exist specifically to guard this extraction.

Goal

Split the monolith into at least three focused modules with clear single responsibilities, without changing public imports or runtime behavior.

Prerequisites

• Monday PR2 merged — runtime validation at JSONL boundary
• Tuesday fixtures merged — tests/test_real_session_fixtures.py dispatch-order tests must pass before and after split
• Wednesday export work optional (imports parse_session only)

git checkout master
git pull
git checkout -b refactor/jsonl-parser-split

Scope

New modules (minimum)

Module	Responsibility
utils/tool_dispatch.py	All _tool_result_pred_*, _tool_result_build_*, _TOOL_RESULT_DISPATCH, _parse_tool_result(); module header documents ordering contract
utils/session_peek.py	quick_session_info() — two-pass metadata peek
utils/jsonl_parser.py (slimmed)	parse_session(), message processors (_process_user/assistant/system/progress), metadata assembly, validation return

Optional:

• utils/jsonl_helpers.py — shared content helpers (_normalize_content, _extract_text, _strip_system_tags, etc.) if needed to avoid cyclic imports between parser and peek modules
• utils/jsonl_reader.py — raw line I/O only if it clarifies boundaries (not required if parse_session stays readable)

Backward compatibility (required)

Re-export from utils/jsonl_parser.py so existing imports unchanged:

from utils.jsonl_parser import parse_session          # api/, scripts/
from utils.jsonl_parser import quick_session_info     # api/projects.py
from utils.jsonl_parser import _parse_tool_result     # tests/
from utils.jsonl_parser import _TOOL_RESULT_DISPATCH  # Tuesday fixture tests
from utils.jsonl_parser import _strip_system_tags     # utils/md_exporter.py

No test file edits — if tests pass unmodified, re-exports are correct.

Constraints

• No behavior change — mechanical move only; dispatch predicate order unchanged
• No cyclic imports between new modules
• No public API signature changes on parse_session() or quick_session_info()
• Update docs/architecture.md dispatch location only if it references the old single-file layout (one-line touch acceptable)

Out of Scope

• New tool-result types or dispatch reordering
• Runtime validation changes (utils/validation.py)
• New fixtures or parser behavior changes
• Export engine changes (Wednesday)
• Broad documentation rewrite

Acceptance Criteria

• utils/jsonl_parser.py split into ≥3 modules with clear single responsibilities (tool_dispatch.py, session_peek.py, slimmed jsonl_parser.py)
• Dispatch table isolated in its own module with explicit ordering-constraint documentation
• No public API changes: existing imports from utils.jsonl_parser continue to work via re-exports
• All existing parser tests pass without modification (tests/test_jsonl_parser.py)
• Tuesday fixture tests pass without modification (tests/test_real_session_fixtures.py)
• No cyclic imports introduced
• mypy --strict passes on utils (and dependent packages)
• pytest -q green in CI
• PR approved by at least 1 reviewer