blktrace: fix trace mutex deadlock

A previous commit changed the locking around registration/cleanup,
but direct callers of blk_trace_remove() were missed. This means
that if we hit the error path in setup, we will deadlock on
attempting to re-acquire the queue trace mutex.

Fixes: 1f2cac10 ("blktrace: fix unlocked access to init/start-stop/teardown")
Signed-off-by: Jens Axboe <axboe@kernel.dk>