Cephmount thread pool followups - #5969

The cleanup stopped a thread after 5 minutes without activity, also
while a request (for example a large upload) still ran on it. The
response of that request could then be lost, and the requests in the
queue never ran, so their callers waited until their deadline.

The pool now counts the callers of each thread, and the cleanup stops
only threads without callers. This also removes the active flag, which
was written without a lock (data race). A thread in the pool map is
always running, because the pool stops a thread and removes it from the
map under the same lock.

Assisted-by: Claude:claude-opus-5-5

Upload copied the request body to the file on the user thread. The
speed of the client decided how long this took, and during that time
all other requests of the user waited. All external accounts and
public-link guests use one thread, thus one slow upload blocked all of
them.

The file is now opened on the user thread, which creates it with the
user as owner and does the permission check. The data is copied outside
the thread, because a write to an open file is not checked again.
Download already works like this.

The error from Close is now returned. CephFS can report write errors
only when it flushes the data at close, and before this change such an
upload reported success.

Assisted-by: Claude:claude-opus-5-5