backupccl: allow table renames during restore

[shermanCRL]

It is a requirement that, when restoring tables into a database, tables (and views) of those names must not exist in the target database. Besides being inconvenient, it can lead users down a surprisingly frustrating road.

Users would like to do these table restores “in place”, next to the existing tables. We should enable this by allowing renaming the incoming tables at restore time.

Possible syntax using a new AS clause:

RESTORE TABLE bank.customers AS customers2, bank.accounts AS accounts2 
FROM LATEST IN 's3://{bucket_name}?AWS_ACCESS_KEY_ID={key_id}&AWS_SECRET_ACCESS_KEY={access_key}';

After completion, users will end up with two tables. They then have the normal SQL primitives they need to merge (UPSERT etc).

While we are in the neighborhood, should we offer AS for restoring databases as well? This would be semantically equivalent to WITH new_db_name.

@dt Also mentions the idea of a “swap” option -- the existing identically-named table stays online during restore, which may take a while. Upon restore completion, the newly-restored table replaces the previous table, presumably atomically.

See also #55791

See also this Docs issue (internal) to be more prescriptive on when one should restore a table vs a database.

Jira issue: CRDB-24353

[blathers-crl]

cc @cockroachdb/disaster-recovery

[shermanCRL]

(Adding Schema as I am not sure who would need to implement this.)

[ajwerner]

This is inside of restore planning. It uses schema APIs, but this is @cockroachdb/disaster-recovery's to implement.

[ajwerner]

As a note, I am working on making it so that renaming or creating a table with a name that is actively being queried is possible (see #96840), but I imagine this issue remains useful in more contexts.