End the MySQL thread on every poll_host exit

[somethingwithproof]

Closes #594. Stacked on #589; merge that first.

poll_host() leaves through three places. Two called mysql_thread_end(), one did not:

poller.c:810   mysql_thread_end();     /* set.ping_only exit */
poller.c:1962  mysql_thread_end();     /* normal exit        */
poller.c:523   return;                 /* device row missing */

There is no mysql_thread_init() in the tree, so the client library sets its thread-local state up on first use and mysql_thread_end() is what releases it. poll_host() has already issued db_query() before it reaches that branch, so the state exists and is then abandoned. Spine is thread-per-device, so it is once per affected device per cycle.

The branch fires when SELECT ... FROM host WHERE id = %i AND deleted = '' does not return exactly one row: a device deleted or soft-deleted between the poller picking it up and the thread reading it, or a remote poller whose host table has not caught up.

Why it happened

The teardown was written out three times, six SPINE_FREE calls plus the same fourteen-line connection release. The three copies were not in the same order and, as here, did not contain the same steps. Nothing put them side by side.

The change

poll_host_release() for the two early exits, taking pointers so the caller's pointers are nulled rather than left dangling.

The normal exit could not use it: it calls buffer_output_errors(error_string, ...) after the point where the helper frees error_string. Rather than reorder that, it shares poll_host_release_connections() only and keeps the rest of its sequence. So the fourteen-line connection block is now in one place across all three, and the six frees are shared across the two that can share them.

poll_host() drops from 1,620 to 1,570 lines.

Verification

This one is not unit-testable: mysql_thread_end() is a libmysql call with no observable return, and reaching the branch needs a device row to vanish between two queries. Saying so rather than dressing up a weaker test as proof.

What was checked: every return in poll_host(), and its fall-off-the-end exit, now reaches mysql_thread_end(). Clean ubuntu:24.04 build matching ci.yml, same four warnings as develop, make check green on all four binaries, 49 of 49 tests.

No CHANGELOG entry here; #578 is the changelog PR for this batch.

[somethingwithproof]

Consolidated into #597, which carries this branch's commits unchanged.

Every pair of these ten conflicted on tests/unit/test_linked.c because each appended to the same registration array, so whichever merged first would have forced a rebase on the other nine. #597 is one review and one approval for the same 20 commits, still one logical change each.

Nothing here is dropped. Reopen this if you would rather review it separately.