emrg: durable evolution count — include valid disk logs after daemon restart

[pm25coder]

Follow-up to #558 (evolution count aggregation): the count was still inconsistent across daemon restarts.

[argszero]

✅ LGTM — cycle

The fix correctly addresses the #558 gap: the scheduler's in-memory count resets to 0 on daemon restart while evolution-*.json logs persist. Counting valid disk logs (parseable JSON with a timestamp) and taking max(in_memory, disk) keeps the GUI growth card/toast consistent with the 'recent' list across restarts.

Verified:

• Production logs are always top-level dicts with a 'timestamp' key (scheduler._write_evolution_log writes json.dumps of a dict), so the .get('timestamp') guard is sound.
• Positive/negative states covered by tests: valid disk logs counted (count==2 / count==25), corrupt/partial writes skipped (JSONDecodeError caught), and the legacy fallback path (no scheduler) still returns len(self.evolutions).
• Test CI green (31230683979).

[argszero]

✅ LGTM — cycle

Second review pass on the same head (4db7440). Re-verified:

• The count now survives daemon restarts (max of in-memory scheduler count and valid disk evolution-*.json files), keeping GUI growth card/toast consistent with the 'recent' list.
• Corrupt/partial JSON writes are skipped; the legacy fallback (no scheduler) still works.
• Test CI green (31230683979) and the PR is mergeable.

[argszero]

✅ LGTM — cycle

Third review pass (head unchanged, 4db7440). The durable evolution count fix is sound:

• Count = max(in-memory scheduler count, valid disk evolution-*.json files) survives daemon restarts.
• Corrupt/partial writes skipped; legacy fallback intact.
• Test CI green, mergeable. Ready to merge.