opfs: enable opfs temp_directory via pre-registered pool

[dt]

Previously attempting to set temp_directory to an opfs:// path did not work.

Specifically, while 'auto' file handling scanned SQL stmts for opfs:// paths to pre-open mentioned files via the async OPFS API and register them where the sync openFile call could find them, it only worked for specific files and not for a temp directory, i.e. a path in which openFile would later attempt to create individual files.

The sync vs async APIs of openFile vs OPFS presents a challenge here: when the DB decides to spill to a tempfile and attempts to create it, it does so via the openFile call, which is a sync call. This makes creating the OPFS file at that point, via the async OPFS API, a problem.

To get around this, as of this change, when an OPFS temp directory is registered, it sets up a 'pool' of pre-created empty temp files, with sync access handles to each pre-created, which can then be handed out as needed to (sync) openFile calls. When closed, a file can be truncated and returned to the pool, or be deleted if the pool is full. When the pool runs low on pre-created files, new files can be opened -- async -- and added to it.

This somewhat indirect approach works around the sync vs async API mismatch, while still allowing openFile calls to create arbitrarily named -- from its point of view -- files at arbitrary times. The included tests show this in action for opfs temp directories, both when configured via SET at interactively and via up-front configuration passed to open().

Fixes #2061.