pgcolumnar.truncate() takes its lock and names the relation type before checking ownership (the #749 pattern) #1384

Found while reviewing #1381.

What happens

pgcolumnar_truncate (src/columnar_vacuum.c, around line 2917 on main 9b3c181) does three things in this order:

1. table_open(relid, ShareUpdateExclusiveLock);
2. refuses a non-columnar relation with "relation … is not a columnar table";
3. PgColumnarRequireTableOwner(rel).

This is the order #749 fixed in the five maintenance entry points. It has two consequences:

• A non-owner queues for the lock. If the owner holds a conflicting lock, the non-owner waits behind it, and so does every later request.
• A non-owner learns the relation type of a table they have no rights on.

Reproduced (by the #1381 review): a probe in a copy of vacuum_lock_privilege.sh gave
FAIL REVIEW PROBE truncate refuses a non-owner before its lock: got [locktimeout] want [owner]

None of the privilege suites covers truncate.

Proposed fix

Do what #749 did:

• check ownership by OID (PgColumnarRequireTableOwnerByOid(relid)) before table_open;
• then check the type.

The arms, red first, in vacuum_lock_privilege (both twins):

• "truncate refuses a non-owner before taking its lock", under the existing contention holder;
• "truncate tells a non-owner nothing about the relation type", as notowner_heap does for compact.