When starting tlssh we connect its stdout to a pipe, but we only read from it
during startup. This means that it can hang if too much output is generated
later on. We should make sure we constantly read from the pipe and put the data
in the log.
This is non-trivial because of the way pipes work in Windows (you cannot do
select or poll them).
You can poll them actually. And this is properly done as part of bug 3183.
I've made ssh spew as much data as possible into the logs using "tlclient.exe -d5" and it seems to work alright.