Snapshot patch can produce multi-MB parts with 100K+ file entries

[BYK]

Description

Snapshot.patch() runs git diff --name-only with no file count cap. When a worktree has unignored test artifacts (e.g. .test-tmp with 99K+ files from test runs), the diff returns all file paths — stored as a 17 MB JSON patch part in the DB and sent to the browser.

Steps to reproduce

1. Run tests in a project that creates temp files not covered by .gitignore
2. Start a session with snapshot enabled
3. Observe patch parts with thousands of file entries

OpenCode version

latest (dev)

OS

Linux

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [BUG] Snapshot module ignores 'watcher.ignore' config, causing excessive disk usage and performance degradation #8887: Also about the snapshot system tracking/processing files it shouldn't (large unignored directories) — similar root cause around unbounded file handling in snapshots, though the specific mechanism differs (this issue is about git diff --name-only in Snapshot.patch(), while [BUG] Snapshot module ignores 'watcher.ignore' config, causing excessive disk usage and performance degradation #8887 is about git add in Snapshot.track()).
• Warn or block snapshots when the detected git root is an accidental top-level repo #18857: Related — about the snapshot system operating on unexpectedly large git worktrees with many files.

If your issue is specifically about the patch part size from Snapshot.patch(), it may be a new and distinct problem worth tracking separately.