Previously for incoming transports, the diagnostic_name (for log messages)
was just "responder" -- not sufficient to distinguish peers.
We now use the pubkey instead.
For outgoing transports it is f"{host}:{port}" (unchanged).
We could just use the pubkey for both uniformly; but it is quite long, and
it is hard to distinguish them at a glance.
Scenario (prior this change):
A task in lnpeer.Peer.taskgroup raises ORIG_EXC, e.g. in htlc_switch.
The taskgroup then cancels all its tasks and then awaits each (in cancel_remaining):
4e64c56042/aiorpcx/curio.py (L217-L221)
In Peer.main_loop, we would want ORIG_EXC to be raised,
but instead LightningPeerConnectionClosed() will be raised as
the Peer._message_loop() task is cancelled, and it is awaited first in cancel_remaining.
We should make sure that if a task is cancelled it will let the CancelledError
propagate out, or at least it does not raise a different exception instead.
In recv_rev() previously all unacked_local_updates were deleted
as it was assumed that all of them have been acked at that point by
the revoke_and_ack itself. However this is not necessarily the case:
see new test case.
renamed log['unacked_local_updates'] to log['unacked_local_updates2']
to avoid breaking existing wallet files
pycryptodomex 3.7 implemented chacha20_poly1305 and chacha20,
and it is already used (although optionally) to speed up AES,
so we can remove cryptography and make pycryptodomex mandatory for LN
