feat: add Router Memory diagnostics and fail-open runtime receipts

Summary

Add production diagnostics and fail-open receipts for Router Memory usage in agent routing.

Parent roadmap: #2287 Parent issue: #2339 Related: #2239, #2240, #2243, #2338 Related code: src/semantic-router/pkg/memory, src/semantic-router/pkg/extproc, src/semantic-router/pkg/routerreplay, src/semantic-router/pkg/observability, src/semantic-router/pkg/sessiontelemetry

Motivation

Router Memory can affect multi-turn and agent routing decisions, but production users need to know whether memory was used, stale, unavailable, ignored, or blocked by policy. The current roadmap has runtime, materialization, and distributed-state tracks; this issue focuses on operator-visible receipts.

Scope

Implement diagnostics for:

• memory lookup/extraction/rewrite status;
• memory backend availability and freshness;
• fail-open/fail-closed reason;
• interaction with safe exploration/protection policies;
• replay fields and metrics for memory-influenced routing.

Non-goals

• Do not add a new memory algorithm here.
• Do not store raw sensitive memory content in aggregate diagnostics.
• Do not require synchronous memory lookup as a hard dependency for every request.

Acceptance criteria

• Replay can explain memory status for each routed request.
• Metrics distinguish memory used, stale, missing, disabled, and policy-blocked states.
• Missing memory degrades according to the documented runtime contract.