Discard stale WebSocket state when reconnecting

[bsergean]

When a connection drops during a fragmented receive or a backpressured send, the transport retains state from that session. A reconnect can combine OLD from the previous connection with NEW from the next, or send an old frame tail before the next message.

Reset queued output, incomplete receive data, receive targets and fragment assembly state before a new client or server handshake. Finish any active application send before resetting and replacing the transport, so it cannot append old continuation frames after the reset. The wait remains cancellable so a send progress callback can call stop() without deadlocking. Release the write lock before Open callbacks so those callbacks can send normally.

Queued bytes belong to their original connection and are discarded on reconnect; this does not add automatic message replay.

Validation:

• Six local test executables passed: reconnect, server, close, subprotocol, send timeout, and native Network.framework TLS.
• New loopback tests cover automatic reconnects, incomplete frames, queued output, failed reconnect attempts an interrupted fragmented send, and stopping/restarting from its progress callback.
• Removing the reset makes all four regression cases fail. Removing only send coordination makes the interrupted-send case fail independently.
• Replacing the cancellable wait with a blocking lock reproduces a deadlock in the progress-callback stop test.
• New test formatting and git diff --check passed.

Fixes #610.
Fixes #611.