v0.2.15

• Fix read_timeout losing data, EOF, or error notifications when the transport callback
  runs before the read waiter is registered, notably on Python 3.9/3.10 (#155).
• Close the in-flight connection when Pool.wait_closed() is cancelled, and serialize
  concurrent drainers so shutdown cannot report completion before connections close (#156).
• Raise DBAPI OperationalError for writes to closed transports (including uvloop) so
  SQLAlchemy's pool_pre_ping can recognize the disconnect and replace the connection (#158).
• Fix Pool.wait_closed() blocking forever when the last connection was released while inside a
  transaction or after disconnecting, and notify pool waiters when a pool-filling connect()
  raises (#154).
• Fix Pool.terminate() not notifying wait_closed() waiters parked on the pool's condition, so
  a wait_closed() call already in flight when terminate() runs would hang forever even though
  the pool was fully drained (#157).

Full changelog: v0.2.14...v0.2.15