fix gemini tokenUsage in metadata.json for manual commit strategy

[peyton-alt]

Note

Medium Risk
Touches condensation logic that writes checkpoint metadata and changes transcript parsing/token accounting paths for multiple agent formats; mistakes could skew stored metadata or summaries but do not affect core git history or authentication/security.

Overview
Manual-commit condensation now threads CheckpointTranscriptStart into extractSessionData and uses it to compute checkpoint-scoped token usage for both Claude (JSONL line offsets) and Gemini (JSON message indexes).

Transcript processing is refactored to centralize format-aware counting (countTranscriptItems), prompt extraction (extractUserPrompts with IDE-tag stripping), and token calculation (calculateTokenUsage), and commit summarization avoids line-slicing for Gemini transcripts to prevent malformed input.

Tests are updated and expanded with new Gemini condensation coverage, including a multi-checkpoint regression test that verifies token usage only counts messages after the checkpoint start.

^{Written by Cursor Bugbot for commit 8bfa341. This will update automatically on new commits. Configure here.}

[Copilot]

Pull request overview

This PR fixes missing Gemini token usage calculation in the manual commit strategy's metadata.json. Previously, token usage was only calculated for Claude Code transcripts (JSONL format), but not for Gemini CLI transcripts (JSON format).

Changes:

• Added agent-specific token calculation that detects Gemini JSON format and uses the appropriate parser
• Removed TODO comments indicating the missing Gemini token usage implementation
• Imported the geminicli package to access Gemini-specific token calculation

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}