Add a mechanism to let the write queue drain naturally before
closing the TTY, rather than always losing that data.  There
is a timeout, so it can't wait too long.

Provide missing locking inside acm_wb_is_avail(); it matters
more now.  Note, this presumes an earlier patch was applied,
removing a call to this routine where the lock was held.

Slightly improved diagnostics on write URB completion, so we
can tell when a write URB gets killed and, if so, how much
data it wrote first ... and so that I/O path is normally
silent (and can't much change timings).

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>