[moetemp]

What changed?

This PR adds the Stream ids, offsets, epoch keys and batch digest, with shared vectors.

• An offset packs the Redis entry id <ms>-<seq> as ms << 20 | seq.
• Epoch keys are fixed-width big-endian bytes, so a byte compare orders the epochs. The vectors give them in hex.
• A Producer's kind is the case of its epoch. epoch::ProducerKind gives it, with a fixed number for the stored field names, and refuses a Producer whose epoch has no case.
• testdata/gen_vectors.py is a second implementation. The crate's tests read every vector file, and sdk-python also reads ids.json and withdrawals.json.
• testing::fields, behind the test-support feature, has the stored field names that the store suite checks.
• At the completed task's started id, the withdrawal rule withdraws a Record whose task key is lower than the completed task's. C1 sends that task's attempt and started time, and the vectors use them. A completed task with no attempt withdraws nothing at its id.
• The batch digest also hashes whether a Record has a body, and the sizes of its external payloads.
• A Stream has a third key, :producers, the sorted set of each Producer's last append.
• check_producer takes only ids in the form of encode_part.

Depends on #212 (K1). Review lane is the Rust reviewers.
Part of the Brook review series.

Why?

Every SDK must compute these the same way. They have no I/O, so reviewers can check them against the vectors before any Store exists.

How did you test it?

cargo fmt, cargo clippy -- -D warnings (with and without features), cargo test -p temporalio-streams --all-features and cargo doc pass at this commit.

[moetemp]

Closed in favor of #228, the combined branch, which has this change.