bulk: retry transient instance-lookup failures for execution locality across bulk jobs

**Summary:**
`MaybeRelocateJobExecution` marks a failure to resolve instances matching the
execution-locality filter with `ErrRelocateInstanceLookup` (transient — e.g.
nodes still starting) so each caller can set its own retry policy. Only CHANGEFEED
and now IMPORT mark it retryable (`jobs.MarkAsRetryJobError`); BACKUP/RESTORE do
not, so a transient lookup failure fails the job instead of retrying.

**Proposed work:**
- [ ] Apply `MarkAsRetryJobError` on `ErrRelocateInstanceLookup` to BACKUP/RESTORE.
- [x] Add test coverage for the retry branch (currently untested in IMPORT and
CHANGEFEED), likely via a testing knob on `MaybeRelocateJobExecution` that
forces the lookup error.

Epic CRDB-65504

Jira issue: CRDB-66932