Add configurable serialized-payload limits with fail-open oversize behavior

[lan17]

Resolution

Closed as not planned. DialCache will not add a default or required serialized-payload ceiling.

Rationale

• A universal ceiling would be an arbitrary cacheability policy. Existing applications may intentionally cache large values, and an upgrade must not silently turn those entries into sustained source fallbacks.
• DialCache already observes the exact serialized byte size of every Redis write through the backend-neutral size observation, exposed as a Prometheus histogram or Datadog histogram/distribution when a metrics adapter is configured.
• Applications own their value shapes, serializer, Redis capacity, and operational alerts. An application that needs a hard limit can enforce its own policy in a custom serializer; serializer failures already fail open without losing the fallback result.
• Optional thresholded compression remains separately tracked in Add thresholded client-side compression for Redis payloads #60, but it is an optimization rather than a production prerequisite.

Accepted tradeoff

Observability is reactive, and compression cannot protect Redis from a very large incompressible value. That risk is explicitly application/deployment-owned rather than enforced by a new DialCache public policy.

No public API, cache key, Redis frame, Lua protocol, or runtime behavior changes are planned for this issue.