Avoid scanning non-default partitions

[nicolagi]

This is a minimal change to address #660.
It changes one query from selecting the max timestamp in a given partition (which may incur a full table scan) to selecting any timestamp (which should be fast).

Testing. I tested it with:

pg_prove -U $myuser -d $mydb -ovf test/*.sql

Proof. I'll argue that the query change does not change the behavior of the function.

Let P be the partition being iterated on.

If it's empty, then v_max_timestamp evaluates to null in both old and new code, and the loop iteration does nothing in both old and new code.

Let's then consider a non empty partition P.
Let A (for any timestamp) be the timestamp returned by the new code, and let M (for max) be that returned by the old code.
We have A <= M.

If A >= CURRENT_TIMESTAMP then show_partition_name gives back P for both old and new code. (Using A or M doesn't make a difference, as they're both in P.)

Let's look at the case when A < CURRENT_TIMESTAMP.
If also M < CURRENT_TIMESTAMP, both the old and the new code would use CURRENT_TIMESTAMP, so no difference.

It remans to consider the case

A < CURRENT_TIMESTAMP <= M,

in which case CURRENT_TIMESTAMP still belongs to partition P, so the old code (uses M) and the new code (uses CURRENT_TIMESTAMP) both return P.

[keithf4]

Thanks! Will take a look at this when I return to development.