(2.12) Atomic batch: support deduplication

[MauriceVanVeen]

2.12.0 released atomic batch publish, but didn't contain deduplication support yet. Now that we've decided to have both atomic batch publishing (atomic & consistent) and non-atomic high-speed publishing (performance), we can make a decision around how deduplication works as well.

When the Nats-Msg-Id deduplication header is used and a duplicate message is detected, the whole batch gets rejected with an error: atomic publish batch contains duplicate message id. This ensures the all-or-nothing-gets-persisted atomic nature is preserved, also when deduplication is used. There are no exceptions.

The high-speed non-atomic batching could then support deduplication differently by allowing them to be omitted from the stored messages, similar to how js.PublishAsync would work.

ADR-50 update: nats-io/nats-architecture-and-design#385

Resolves #6980

Signed-off-by: Maurice van Veen github@mauricevanveen.com

[neilalexander]

LGTM