mount: Wrap yields in context managers with try-finally

The terminator context will only clean up on a signal,
so if another exception causes context manager cleanup
then unmount won't be called unless it's part of another context manager
or is wrapped in a try block's except or finally.

Everywhere else's unmounts are handled by delegating to another context manager
but these were what needed to be fixed.

The change in buildstream/_fuse/mount.py would cause lockups
since the build worker process still having a subprocess
blocked its termination from completing
which in turn caused the pipeline manager process to block indefinitely on it.