sql: stabilize test expectations wrt schema descriptor IDs

Currently every time we add a new object in the system database, it shifts the start descriptor ID for all the non-system tables and other SQL objects. Then all the test expectations throughout the code base become invalid.

So the work needed to add a new object in the system database is abnormally complex/excessive, as the person doing this needs to revisit every other test in many areas they are not concerned with.

The solution that was discussed with @rafiss and @fqazi is to bump the descriptor ID sequence in **test servers** so that the first SQL object defined by tests starts at a stable ID independent of how many objects are included in the system database.

An example way to do this is this:
```go
ie := tenant.InternalExecutor().(*sql.InternalExecutor)
_, err := ie.ExecEx(⋄, "force-desc-id", nil, /* txn */
sessiondata.NodeUserSessionDataOverride,
`SELECT setval('system.descriptor_id_seq', 10000)`)
require.NoError(t, err)
```

We need special consideration (to be documented as comment in the code) about the choice of 10000 vs another integer offset.

We need a value such that, for many IDs generated sequentially from it, we preserve the nice property that `MakeTablePrefix(ID).PrefixEnd() == MakeTablePrefix(ID+1)`.

This is true of all values between 1 and 108, but e.g not true of 109, 2xx, 511, 7xx, 1023.

So for example, 1000 would be a poor offset choice because as soon as 23 SQL objects are defined, we'd get artifacts in test expectations (e.g. "/PrefixEnd" appears in key string representations).

10000 is better because the first value after 10000 that doesn't work is 10239, so we get 239 IDs that behave peacefully.

Jira issue: CRDB-31457
Epic: CRDB-28893

Locate the test-server setup and the code that initializes or advances system.descriptor_id_seq, then review the InternalExecutor example in the issue. The change is complete when SQL objects created by tests start from a stable offset, the offset rationale is documented in code, and affected test expectations remain stable as system objects change.

Written by the indexing model from the issue text.