[jsntcheng]

Closing a StunProtocol while a transaction is in flight leaves the transaction running: its retry timer keeps calling send_stun() on the closed transport, and the awaiting request() only finishes once all retransmissions time out (~40 s with the defaults). In aiortc this shows up as errors logged after RTCPeerConnection.close() when ICE checks were still in progress.

This adds Transaction.abort(), which cancels the retry timer and fails the transaction with a new TransactionAborted (a TransactionError, so existing except stun.TransactionError handlers in check_start keep working). StunProtocol.connection_lost() aborts all pending transactions.

Tests: test_abort (no retransmission after abort, abort is idempotent) and test_close_aborts_transactions (closing the protocol fails the pending request). Both fail without the change.