Fix: AsyncTunnelHTTPConnection leaks connection on handshake failure

[drshvik]

Summary

Fixes #1023.

When using an HTTP proxy with tunneling (HTTPS requests), if the CONNECT handshake fails (e.g. RemoteProtocolError, timeouts, or TLS errors during the tunnel setup), the underlying network stream was not explicitly closed.

This caused the connection to remain in the pool as "active" but dead. If max_connections is set, this eventually leads to pool starvation where no new requests can be made.

The Fix

• Wrapped the connection setup logic in AsyncTunnelHTTPConnection.handle_async_request with a try...except block.
• Explicitly calls await self._connection.aclose() if any exception occurs during the handshake or TLS upgrade steps.

Verification

I verified this with a local reproduction script that simulates a proxy server accepting a connection and then immediately closing it (crashing the handshake).

Before the fix:

• The pool count remained at 1 (stuck connection).
• Subsequent requests hung indefinitely waiting for a pool slot (if max_connections=1).

After the fix:

• The pool count correctly returns to 0.
• Subsequent requests fail immediately (correct behavior) instead of hanging.

[drshvik]

This relates to #1049 but provides a broader fix by wrapping the entire handshake process, ensuring cleanup even if the failure occurs before the TLS upgrade step.