Digest 93a2929d, commit 44820d4. Live on all three pods after a staged 2->1->0
roll with a POST /items quorum probe between stages (201 each).
The roll needed ZERO manual pod restarts — the previous one needed three because
two reseed markers latched per node and only one discharged. Pods came Ready in
30s / 80s / 110s.
Verified against the exact probe that caused the outage: 128-dim vector into the
1536-dim content_vector slot now returns 400 with a specific message instead of
500, no receiver halted on any pod, all 9 shard frontiers at lag 0, reseed_gaps
empty, writes 201. Playwright 34/34, semantics 5/5, verify-live 23/23.