Both, the (internal) process/protocol buffer and the log buffer are now
unibyte. The conversion to multibyte UTF-8 is only done for user visible
(UI) buffers.

To properly handle the protocol line termination (CRLF), I added
`:coding 'raw-text-unix` (with explicit unix EOL convention) to the
`open-network-stream' call (also in the new `manage-sieve-encode'
function. This was needed to allow keep the various (looking-at
"...\r\n" ...) calls working. This is something which still feels a bit
weird, but I haven't found another way to do it. I did some tests with
(setq-default buffer-file-coding-system 'utf-8-unix/'utf-8-dos) which
did not show any issues.

I would also add some ERT tests, probably in a separate commit?


In addition, I found that `sieve-manage-quit' in sieve.el had the
tendency to kill unrelated buffers in case of errors during earlier
steps. For this, I created a sepate patch: