sql/catalog/tabledesc: fixed a bug in `maybeAddConstraintIDs`

[Xiang-Gu]

Previously, when we RunPostDeserializationChanges, one of the step is to fill in constraint IDs if they are not already set, which is done in function maybeAddConstraintIDs. That function however is buggy in that, if there is constraint mutation on the table descriptor, we directly go assign a constraint ID (from tbl.nextConstraintID) to it, without checking whether it already has a non-zero constraint ID or not. This PR fixes that.

Release note: None

[cockroach-teamcity]

This change is 

[Xiang-Gu]

TFTR!

bors r+

[craig[bot]]

Build failed (retrying...):

• Bazel Essential CI (Cockroach)

[craig[bot]]

Build failed (retrying...):

• Bazel Essential CI (Cockroach)

[craig[bot]]

Build failed:

• Bazel Essential CI (Cockroach)

[Xiang-Gu]

The merge failed due to a flaky test, retrying

bors r+

[craig[bot]]

Build failed (retrying...):

• Bazel Essential CI (Cockroach)

[craig[bot]]

Build failed:

• Bazel Essential CI (Cockroach)

[Xiang-Gu]

It's sad to see merge failure again but it seems unrelated to this change, retrying again, bors r+

[craig[bot]]

Build succeeded:

• Bazel Essential CI (Cockroach)