Design batch reads and invalidations without one Redis script per key

[lan17]

Priority

P3 — Low — later or demand-driven enhancement.

Problem

The public wrapper and semantic Redis boundary are scalar. Fetching or invalidating many IDs therefore causes one network/script operation per key even when the source of truth supports batching.

Evidence:

• Scalar cached wrapper
• Scalar Redis read/write/invalidate interface

Open design

Produce a typed API proposal and coordinate it with the cache-definition-handle work in #21 before implementation.

Acceptance criteria

• Define loader behavior, ordering, misses, partial failures, and fail-open semantics.
• Preserve runtime policy, ramping, local cache, coalescing, serializers, and tracked invalidation.
• Group Redis Cluster operations by slot and handle tracked/untracked keys correctly.
• Define a capability fallback for scalar-only custom adapters.
• Benchmark batches of 10, 100, and 1,000 against scalar Promise.all.
• Obtain explicit API approval before implementation.