[ENHANCEMENT] Checkpoints: respect .rooignore and add cleanup/size limits

[hannesrudolph]

Type

Enhancement

Problem

Checkpoint storage can grow extremely large when projects include big/binary assets. Users report hundreds of GB ballooning to TBs within a short time, with no automatic cleanup or hard limits.

Context

• Affects repos with large binaries, media, datasets, or LFS-tracked files.
• The checkpoint mirror does not currently respect a user’s .rooignore, so files the assistant will never edit still get mirrored.
• There is no automatic retention or size cap for old checkpoints, so disk usage can continually increase during active work.
• Users don’t want to change .gitignore just for Roo; they prefer using .rooignore to exclude non-editable/binary content from checkpoints.

Desired Behavior

• Respect .rooignore for checkpoint storage so non-editable/binary content is excluded by default.
• Provide a simple cleanup policy:
  • Time-based retention (e.g., keep last N days).
  • Max checkpoints per task.
  • Global size cap for checkpoint storage with oldest-first cleanup.
• Offer a mode that avoids capturing unchanged or non-relevant large files (e.g., “modified files only” behavior) so snapshots focus on actual work.
• Optional lightweight approach to avoid pulling in big directories (e.g., sparse-like behavior) for users with very large trees.
• Add a manual “Clean up old checkpoints” action and safe defaults to prevent disk exhaustion without user intervention.

Reproduction

1. Open a project containing large/binary or LFS-tracked files.
2. Use the extension with checkpoints enabled and perform several task iterations.
3. Observe disk usage growth from the checkpoint storage.
4. There’s no built-in way to set retention limits or exclude non-editable binaries via .rooignore.

Constraints/Preferences

• Do not require changes to .gitignore; rely on .rooignore for Roo-specific exclusions.
• Safe defaults that prevent disk exhaustion.
• Keep the flow simple and unobtrusive for users who don’t need fine-grained control.

[roomote]

I'm exploring a potential fix for the checkpoint storage enhancement. I'll work on implementing .rooignore support and adding cleanup/size limit policies to prevent disk exhaustion.

[roomote]

I've successfully implemented the checkpoint enhancements and opened PR #8042: #8042

The implementation includes:
✅ .rooignore support for checkpoint storage
✅ Time-based retention policies (configurable, default 7 days)
✅ Count-based limits per task (configurable, default 50 checkpoints)
✅ Global size caps with automatic cleanup (configurable, default 5GB)
✅ Manual cleanup capability
✅ Comprehensive test coverage

All CI checks are passing. The solution addresses the disk exhaustion issue while respecting user preferences through .rooignore. Feedback welcome!