• rename the drain routine to ath_tx_tid_drain(), bringing it (mostly) in line with what the reference/atheros driver code does.

• add some locking around the sequence number assignment, since although only one location allocates sequence numbers, now we're changing it in >1 location.

This is inspired by the linux/atheros reference code, but as mentioned in
the comments in this commit, there's a possible race where an interface
is flushed whilst active traffic is being queued to it (or being completed.)
This may cause some funny shenanigans to occur.

This doesn't happen (as often) in the Linux/Reference code because the TID/TXQ
locks are being held for much longer. I'll have to fix this for access point
mode as otherwise some crazy stuff is likely to occur.