self.encoding which we set in the init is only intended
to decode gpg´s stderr which uses a system specific encoding.

if we dont encode the data we pass to python-gnupg ourself, it will fallback and use self.encoding.
This might be of no concern if self.encoding is set to 'utf8' and when we are on Linux
which has a preferred encoding of 'utf8'.

But if we are on Windows the preferred encoding for stderr
is most of the time not 'utf8'. If python-gnupg tries to decode a stderr stream that is for example
encoded with 'cp1252' with our set encoding of 'utf8' this will fail.

The solution is to pre-encode the data before we pass it to python-gnupg, so it does not have to
use self.encoding as a fallback. And set self.encoding='latin1' because latin1 will not yield exceptions
on decoding errors. Also gpg itself will fallback to latin1 as stderr encoding when it cant determine the
preferred encoding of a system.

self.decode_errors is used for something differently, and has no influence on the situation.

Fixes #8644