Add cache proxy origin metrics

[bill-ph]

Summary

• add Prometheus metrics for cacheable origin fetch outcomes, retry reasons, and active in-flight origin fills
• keep request behavior unchanged; this PR does not add local admission control, throttling, or backpressure
• add regression coverage that concurrent distinct origin misses are not locally rejected

Compatibility

• rebased onto feat(cache-proxy): emit standalone OTLP traces #790; these metrics layer on top of the merged cache-proxy tracing spans without copying or changing that tracing work

Follow-up

• add the new origin fetch outcome/retry/in-flight metrics to grafana-dashboards/managed-warehouse/duckgres-cache-proxy.json
• bounded proxy backpressure/queueing remains a separate TODO/PR
• investigate singleflight leader-cancel behavior for same-key origin fills; today the shared fill uses the first request context, so a leader disconnect can affect other waiters

Validation

• go test -v ./cmd/cache-proxy
• focused metric/concurrency tests with -count=50
• focused metric/concurrency tests with -race -count=20
• just test-unit
• just lint