fix(digest): preserve earlier digest when re-run scans zero sessions

[Joi]

Summary

A mid-day re-run that finds only already-processed sessions would overwrite the morning's populated digest with an empty one, because the in-memory signal lists only contain this run's findings — not the prior run's.

Skip the write when sessions_scanned == 0 and the target digest file already exists. The first run of the day still produces an empty skeleton, so downstream tooling can grep for today's file unconditionally.

Why merging is the conservative choice

The signal vectors (all_corrections, all_errors, …) are populated only from sessions scanned in the current call. There's no way to merge today's earlier signals with this run's zero signals without reloading the prior digest, parsing it, and re-rendering — and the prior digest is already byte-stable downstream-grep input. Preserving the file is strictly safer than synthesizing a partial overwrite.

Test plan

• run_review_preserves_existing_digest_when_nothing_scanned — seeds a digest, runs run_review with zero readers on the same date, asserts the seeded SEEDED-FROM-EARLIER-RUN marker survives.
• run_review_writes_digest_when_file_absent_even_with_no_sessions — asserts the first-of-day empty-skeleton path still writes a fresh digest with signals_captured: 0.
• All 85 jilog-review tests pass.
• CI green.