sql: disallow partial unique constraints in ALTER TABLE

[mgartner]

Previously, a WHERE clause in ALTER TABLE .. ADD CONSTRAINT .. UNIQUE
statements was parsed successfully but ignored so that the resulting
unique constraint was not partial. This commit disallows the WHERE
clause. Postgres allows partial unique indexes, but not partial unique
constraints, so we will likely remove support for partial unique
constraints in CREATE TABLE statements in an upcoming major release.
See discussion in #65825.

Fixes #67234

Release note (bug fix/sql change): A bug was identified that created
non-partial unique constraints when a user attempted to create a partial
unique constraint in ALTER TABLE statements. Creating partial unique
constraints in ALTER TABLE is now disallowed. A partial unique index
can be created instead with CREATE UNIQUE INDEX .. WHERE.

[cockroach-teamcity]

This change is 

[ajwerner]

I'm cool with this but we should get @vy-ton to sign off. The other option is to fix it, issue a NOTICE that we may remove it, and then treat it as though it was a CREATE INDEX. The fact that we support partial UNIQUE WITHOUT INDEX makes the whole thing messy.

Reviewed 2 of 3 files at r1, 1 of 2 files at r2.
Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @mgartner and @michae2)

---

pkg/sql/logictest/testdata/logic_test/alter_table, line 1488 at r2 (raw file):

# Partial predicates are allowed for UNIQUE WITHOUT INDEX, but not UNIQUE.
# TODO(mgartner): This is confusing, and something we should think more about
# if we ever make UNIQUE WITHOUT INDEX available by default.

Do we like this?

[vy-ton]

I'm inclined to leave CREATE TABLE as is until we determine our long term plan for unique index vs constraint

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @mgartner and @michae2)

[michae2]

but I defer to @ajwerner and @vy-ton

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @mgartner)

[mgartner]

I'm inclined to leave CREATE TABLE as is until we determine our long term plan for unique index vs constraint

That's fair. So should we just allow predicates it in ALTER TABLE for now then and call it a day?

[ajwerner]

I'm inclined to leave CREATE TABLE as is until we determine our long term plan for unique index vs constraint

That's fair. So should we just allow predicates it in ALTER TABLE for now then and call it a day?

Yeah. I think there's a question of whether we should mark the resultant index as created_explicitly or not.

[mgartner]

Closing in favor of #68629.