fix: preserve UTF-8 calendar snippets during sync and repair

[danshapiro]

Calendar sync currently caps previews by slicing the first 200 bytes. When byte 200 falls inside a multi-byte character, the stored snippet is invalid UTF-8 and can make analytics cache reads fail.

This keeps the existing 200-byte calendar preview limit while backing up to a valid character boundary. It also extends the existing repair-encoding path to faithfully regenerate only the exact historical damage signature: an invalid 200-byte calendar snippet that is a byte-for-byte prefix of a present, valid canonical body. Other invalid snippets retain the existing generic encoding repair, and valid snippets are left untouched.

The repair command's existing full cache rebuild republishes corrected previews, so no cache schema or publication changes are needed.

[roborev-ci]

roborev: Combined Review (02b68ed)

Code is clean at the requested severity threshold; no Medium, High, or Critical findings were reported.

---

Reviewers: 2 done | Synthesis: codex, 10s | Total: 3m14s

[roborev-ci]

roborev: Combined Review (392f7f6)

No issues found.

---

Reviewers: 2 done | Synthesis: codex | Total: 3m25s