lntransport: optimise read_messages implementation

Not great to use a 'bytes' object as a FIFO buffer, as every slice copies the whole thing.
With bytearray, extending it from the right is fast,
and with the correct syntax, popping from the left is fast too.

see https://stackoverflow.com/a/57748513
https://bugs.python.org/issue19087
https://github.com/python/cpython/commit/5df8a8a1fd6cc6f4469dc7d3994d06e2aea24c52

electrum/lntransport.py

@@ -105,19 +105,19 @@ class LNTransportBase:
         self.writer.write(lc+c)
 
     async def read_messages(self):
-        read_buffer = b''
+        buffer = bytearray()
         while True:
             rn_l, rk_l = self.rn()
             rn_m, rk_m = self.rn()
             while True:
-                if len(read_buffer) >= 18:
-                    lc = read_buffer[:18]
+                if len(buffer) >= 18:
+                    lc = bytes(buffer[:18])
                     l = aead_decrypt(rk_l, rn_l, b'', lc)
                     length = int.from_bytes(l, 'big')
                     offset = 18 + length + 16
-                    if len(read_buffer) >= offset:
-                        c = read_buffer[18:offset]
-                        read_buffer = read_buffer[offset:]
+                    if len(buffer) >= offset:
+                        c = bytes(buffer[18:offset])
+                        del buffer[:offset]  # much faster than: buffer=buffer[offset:]
                         msg = aead_decrypt(rk_m, rn_m, b'', c)
                         yield msg
                         break
@@ -129,7 +129,7 @@ class LNTransportBase:
                     s = None
                 if not s:
                     raise LightningPeerConnectionClosed()
-                read_buffer += s
+                buffer += s
 
     def rn(self):
         o = self._rn, self.rk