rangefeed: if catchup iterator fails to create processor is stopped

[aliher1911]

When rangefeed created it can request a catchup iterator creation from underlying store.

Recent changes to interface can cause creation to fail with error. This failure causes processor to stop immediately.

This behaviour is not correct from rangefeed client perspective. Processor can serve multiple clients at the same time and other clients should not be affected by new feed failure. The better approach should be to fail new registration instead.

There's currently no easy way to expose errors like that, what could be done instead is eager creation of iterator. In that case if iterator creation fails, we don't need to create new registration.

Jira issue: CRDB-31745

[blathers-crl]

cc @cockroachdb/replication

[blathers-crl]

Hi @erikgrinaker, please add branch-* labels to identify which branch(es) this release-blocker affects.

_{🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf.}