importer: add ImportStartWallTime table descriptor field

[msbutler]

This patch introduces the new ImportStartWallTime table descriptor field which
provides the time at which an in-progress import began writing data to disk.

In future PRs, this field will be used to:

• incrementally back up in progress imports, preventing a large BACKUP workload
  after an Import finishes.
• elide importing keys in RESTORE, ensuring a table -- with an in-progress import
  in the back up-- contains no in-progress importing keys and is available on
  the restored cluster.

Informs #85138

Release note: none

[cockroach-teamcity]

This change is 

[msbutler]

@dt I added an ImportType field to the table descriptor. I need this ensure that IMPORT PGDUMP descriptors remain offline once the RESTORE completes.

[dt]

@msbutler Can't we just only set the import wall time field to a non-zero number for non-new imports? And then all other imports would not be restored?

[msbutler]

My motivation for adding the ImportType is that during the restore job, i need to distinguish an IMPORT PGDUMP/MYSQL from any kind of IMPORT INTO, because for the former, I need to keep its table descriptor offline.

For the IMPORT PGDUMP case, instead of adding the ImportType, I considered setting desc.ImportStartWalltime to a special time that will never get hit by IMPORT INTO, like 1. But that seemed a bit confusing for the caller. Happy to go that route if you think that's better/simpler.

[msbutler]

You know what, I'm going to scrap the ImportType. I like the idea of using the ImportWallTime with a special value for IMPORT PGDUMP. It simplifies things.

[dt]

The special value is just... zero, right? And then we only restore offline descriptors if they have a non-zero value, and we elide all rows at/above that value?

[dt]

an IMPORT PGDUMP/MYSQL ... I need to keep its table descriptor offline.

We just don't restore the desc at all here, right? we don't want to keep it offline since there is no job that is going to come along and remove it later.

Basically, it seems like this should be a) if importing into an existing table -- empty or not --, record the time that all import writes will be at/above in the desc b) when restoring, if a table is offline we do not restore the desc or data unless it has a import-writes-are-at-or-above time, in which case we do restore it but elide all rows at/above that time, and the table ends up public at the end of restore just like all the others.

Does that sound right?

[msbutler]

Ah yes, for an in-progress IMPORT PGDUMP, we really should drop all of its associated descriptors on RESTORE.

One piece you're missing: I only backup offline tables iff its ImportWallTime > 0. If we kept Import PGDUMP's wall time == 0, then we can't incrementally back it up. Hence, it needs another special value, perhaps -1.

I wonder if it's worth only focusing on the IMPORT INTO case for now, and revisit incrementally backing up IMPORT PGDUMP in stability/23.1.

[msbutler]

removed the import type. Currently nothing gets set for IMPORT PGDUMP. Would like to merge this ASAP to make it easier to work on downstream stuff.

[msbutler]

TFTR!

bors r=dt

[craig]

Build failed (retrying...):

• Bazel Essential CI (Cockroach)

[craig]

Build failed (retrying...):

• Bazel Essential CI (Cockroach)

[craig]

Build failed (retrying...):

• Bazel Essential CI (Cockroach)

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)