Since much more than the number of bytes we're reading can be sent on a

channel we ignore/close, we would get a separate warning for each chunk of
data on that channel until we read past the close.  Instead, if we're closing
a channel, we note the channel ID, and simply discard the information
immediately on the most recently closed channel.

A remote could still cause a bunch of log output by multiplexing two or more
writes to invalid channels, but nobody seems to do that, so this should be
good for now.