F-antigravity-sessions: Codec / Parser for transcript.jsonl

[bastien-gallay]

Issue #144 — Antigravity CLI Transcript Parser

Add a parser for the Antigravity transcript format to extract session metadata.

Background

Antigravity CLI (agy) logs are saved in a JSONL format (transcript.jsonl)
under ~/.gemini/antigravity-cli/brain/<conversation-id>/.system_generated/logs/.
Unlike Claude Code, which registers events with properties like message,
cwd, and type, Antigravity structures logs as conversation steps.

Requirements

1. Implement a pure codec submodule in crates/claude or a new
   crates/antigravity crate to parse transcript.jsonl.
2. Extract the first USER_INPUT step content as the summary (truncated to
   120 UTF-16 units).
3. Compute message_count based on the number of USER_INPUT and
   PLANNER_RESPONSE steps.
4. Concatenate user/model step content for FTS indexing (text_content),
   capped at 8000 UTF-16 units.
5. Extract the project workspace directory (CWD) from the log (e.g., from step
   metadata or run_command tool call Cwd arguments) to determine the CWD.

Tests

• Parse simple transcripts and verify correct summaries/message counts.
• Gracefully handle corrupt or empty lines (skip them).
• Extract the correct CWD from tool calls or step metadata.

[bastien-gallay]

Refactor tie-in: #170 (E derive.rs seam) — the agy transcript parser plugs in exactly where termherd_claude::derive/digest do today.