ext/pdo_pgsql: avoid getColumnMeta() interfering with ATTR_PREFETCH = 0

[outtersg]

doing an (internal) query to fetch metadata from the server broke the currently-running (user) query

(this is part 2. of #23892 (comment))

[devnexen]

It seems right however, once #23919 is merged, I would like this PR being rebased upon it.

[outtersg]

@devnexen wrote:

It seems right however, once #23919 is merged, I would like this PR being rebased upon it.

I'm totally OK for that, because, even if the second commit is intended to be dropped (containing the part common to both PRs), there will still be the conflict on the intermixed part, about the order to choose for the ifs coming from each PR.

The canonical order being in the original PR:
• let a chance for the cached_table_name to be returned
• then return NULL
• then empty cached_table_name (in order to immediately refill it)

Thus in case of a statement temporarily paused by an unbuffered query's results coming through, we have kept the cache, which can be of use again once the results have been consumed. Strange case but why not.