Peer fetch on local miss with local healing

Summary

If a region misses locally for an object that should exist elsewhere, fetch it from a peer and heal local state in the background or inline, while keeping the read path low-latency.

Why

This gives us a better consumer experience during replication lag, bootstrap windows, or after partial failures without forcing every read through cross-region quorum logic.

Scope

• decide which misses are eligible for peer fetch
• fetch metadata and blob content from a healthy peer
• persist the fetched object locally so later reads stay local
• add timeout, concurrency, and backpressure controls so cross-region healing does not hurt hot traffic
• add metrics for hit/miss, peer-heal latency, and failure rate

Notes

This should be bounded and opportunistic, not a replacement for bootstrap or anti-entropy repair.