[Bug][affinity] Async startup hydration overwrites chains updated by live requests #6253

Reproduction

At master dc041dd1ab1a014efa352ba47a6f31fc47f55284, startup calls hydrateAffinityUnified() in the background. While its KV reads await completion, requests can call prefixAffinity.note() with a deeper or newer chain. Hydration then imports the older file/KV record unconditionally, replacing that live chain.

A controlled test with the real SQLite-backed KV store pauses the first read after capturing a depth-2 snapshot, notes a live depth-6 chain, and releases hydration. The resulting live depth is 2. A second test at equal depth replaces the newer tail hash and prevents that exact replay from retaining its existing session. The one-time legacy migration can also durably write the stale chain over the newer in-memory record.

Proposed fix

Immediately before import, merge the current in-memory snapshot with the file/KV union using the existing deeper-then-fresher persistence rule. Use that reconciled union for one-time migration as well. A genuinely deeper stored chain must continue to win; no new resolution policy is needed.

Validation

• Unchanged master: three new regressions fail; the deeper-stored control passes.
• Fixed: 52 tests pass across unified affinity storage, legacy persistence, prefix resolution and persona restart stability, including actual SQLite and local HTTP coverage.
• Whole-project npm run typecheck, npm run build and git diff --check pass.

Duplicate screening: all existing PR entries, open/closed issues and Afloat16's prior issues/fork branches were checked. #5757 introduced the unified overlay, #5480 concerns cross-process legacy-file journaling, #6047 validates carrier shape, and #6186 concerns write durability/debounce lifecycle. None reconciles mutations made while this startup read is pending.