Fine-grained, keyed job serializability within a workpool with maxParallelism > 1

It would be nice if workpool supported fine-grained (keyed) job serializability based on some job arguments. E.g., I want the workpool have maxParallelism > 1, but jobs that are identical by some (subset of) their arguments are always executed in serializable fashion, at most one at the time.

Internally, this could be implemented by hashing these selected arguments for the given job and assigning the internal thread ID for this job execution to `hashValue % poolSize`, a la https://guava.dev/releases/15.0/api/docs/com/google/common/util/concurrent/Striped.html.

Start by reading the workpool scheduling implementation and its handling of maxParallelism and poolSize. The issue names no files or tests; establish the caller-facing way to select job arguments for a key, then verify that matching jobs run one at a time while different keys can still run in parallel.

Written by the indexing model from the issue text.