Add Redis write-path and end-to-end cache outcome telemetry

[lan17]

Priority

P2 — Medium — closed as not planned after the observability contract was narrowed.

Original proposal

This issue proposed Redis write/client latency, watermark write outcomes, and a broad end-to-end duration/outcome surface in addition to the existing get, fallback, serialization, size, miss, error, and coalescing metrics.

Confirmed scope

The approved production signal is intentionally narrower:

• Keep crisp serializer dump and deserializer load latency.
• Keep serialized Redis value size in exact bytes.
• Preserve request, miss, disabled, bounded error, invalidation, coalescing, get-latency, and fallback-latency metrics.
• Do not add Redis client/write latency merely for completeness.
• Keep the backend-neutral adapter stable unless a demonstrated operational need justifies another public event surface.

That contract is implemented consistently in the core adapter and the Prometheus and Datadog subpaths:

• Backend-neutral metrics interface
• Exact serialization timing and byte measurement
• Prometheus mapping
• Datadog mapping

Future work may add a specific bounded outcome when a concrete approved feature needs it, but should not reopen this broad instrumentation bundle by default.