Fix IP pools data migration

[david-crespo]

Closes #4875

Problem

After the IP pools migrations on the dogfood rack, the default pool was not marked is_default=true for the oxide silo when it should have been.

Diagnosis

When checking for silo-scoped default pools overriding a fleet-scoped default, I neglected to require that the silo-scoped defaults in question were non-deleted. This means that if there was a deleted pool with silo_id=<oxide silo id> and is_default=true, that would be considered an overriding default and leave us with is_default=false on the default pool.

Well, I can't check silo_id and is_default on the pools because those columns have been dropped, but there is a deleted pool called oxide-default that says in the description it was meant as the default pool for only the oxide silo.

oot@[fd00:1122:3344:105::3]:32221/omicron> select * from omicron.public.ip_pool;
                   id                  |        name        |          description           |         time_created          |         time_modified         |         time_deleted          | rcgen
---------------------------------------+--------------------+--------------------------------+-------------------------------+-------------------------------+-------------------------------+--------
  1efa49a2-3f3a-43ab-97ac-d38658069c39 | oxide-default      | oxide silo-only pool - default | 2023-08-31 05:33:00.11079+00  | 2023-08-31 05:33:00.11079+00  | 2023-08-31 06:03:22.426488+00 |     1

I think we can be pretty confident this is what got us.

Fix

Add AND time_deleted IS NULL to the subquery.

Mitigation in existing systems

Already done. Dogfood is the only long-running system where the bad migration ran, and all I had to do there was use the API to set is_default=true for the (default pool, oxide silo) link.

[david-crespo]

These asserts are the same as before, there are just a couple more cases, and I made them denser so I could stare at them more easily for 5 hours.

[david-crespo]

and time_deleted is null is the fix.

[david-crespo]

With this deleted pool added to the test data but the fix commented out, the test fails in the way we expect, which matches the problem on dogfood: the (pool3, silo2) link gets is_default=false instead of is_default=true.