feat: synchronous remote index dropping

[usamoi]

Remote index dropping is asynchronous: the command successes immediately but there will be an error if the handle number is reused by PostgreSQL.

• implement synchronous remote index dropping; related to feat: refactor async task executor #260
• pgvecto.rs uses relNode for remote handle to make reindex concurrently work, but it should use tenant id (32 bits, always set to 0 for now), unique system identifier (64 bits) and oid (32 bits) for remote handle
• use object_access_hook to drop remote indexes, which is more reliable (the dropping implementation does not work for temporary relations and crashed transactions, so it's bad)

Reading materials for contributor:

• Re: Unique cluster / instance identifier?
• GetSystemIdentifier
• object_access_hook
• safe pg_catalog access in pgrx