backport-19.1: sql: ensure lease expiration is monotonically increasing

[vivekmenezes]

Backport 1/1 commits from #36531.

/cc @cockroachdb/release

---

This is a fix for both a correctness issue and a performance problem.

1. Correctness issue during schema changes: A node could hand an
   expiration time from a lease to a transaction T, and if the
   node reacquired a lease for the table at an expiration less than
   the existing lease's expiration it could release the existing
   lease (with a greater expiration) before the transaction T committed,
   thereby invalidating T's deadline.
2. Performance issue: After a lease acquisition if the expiration
   time of the new lease was less than the existing lease, it would replace
   the table in the cache with the new lease and release the older
   lease. But unfortunately it would also not update the table in the
   table name cache (the table name cache always uses the lease with
   the highest expiration). The table in the table name cache would have
   it's lease released rendering the name entry invalid. Therefore until
   the node acquired another lease for the same table, the node would be
   in a state where there was a table cached but it's name -> table cache
   entry invalid, resulting in it making name resolution requests for
   every query (but not lease renewals).

It is likely this bug started happening with the addition of more
aggressive lease acquisitions through #28725. I've reduced the value
of DefaultTableDescriptorLeaseJitterFraction to make the logic
ensuring the monotonicity of the lease expiration fire infrequently.

fixes #36044

Release note: None

[cockroach-teamcity]

This change is