feat: add backend analytics read model for routing quality dashboards

Summary

Add a backend analytics read model/API that aggregates routing quality, distribution, latency, signal confidence, and replay-derived summaries for dashboard consumption.

Parent roadmap: #2287 Parent issue: #2345 Related: #2157, #2200, #2326, #2330, #2333 Related code: dashboard/backend, dashboard/backend/evaluation, dashboard/backend/configprojection, src/semantic-router/pkg/routerreplay, src/semantic-router/pkg/observability

Motivation

Dashboard analytics should not parse raw config or raw replay blobs in the frontend. The backend needs a maintained read model that normalizes route distribution, model selection, latency, signal confidence, and low-confidence/unrouted summaries.

Scope

Implement backend APIs for:

• route/model/recipe distribution;
• latency and overhead summaries;
• signal confidence and projection summaries;
• links to replay detail records;
• redaction/privacy controls;
• query windows and aggregation bounds.

Non-goals

• Do not store raw prompts in aggregate analytics by default.
• Do not invent dashboard-only routing semantics.
• Do not replace #2333 eval reports.

Acceptance criteria

• Frontend dashboard views can consume normalized analytics APIs.
• Aggregates are backed by replay/metrics/config snapshots with redaction.
• API tests cover privacy and summary behavior.