Better string handling for dogwrap and python3 support

[zippolyte]

• Arguments are bytes, so the join must be made on a binary string, otherwise python will try to convert and can encounter errors.
• Stop decoding lines before writing to stdout, since write expects bytes.
• Move decoding the output to the end of the execution, to make sure we won't raise exceptions while command is still running, which can cause unexpected behavior, like in issue Dogwrap doesn't support unicode characters #371 .

Resolves #371, closes #223

[gzussa]

Much better!! and clean description. Thanks for this fix! Should we add tests to make sure it solves the issue?

[synth]

Sorry for the delay in testing, but got to it today. It works! In our testing, dogwrap stays alive and no more database breakdown on unicode output! +1 on adding tests ;-) We'd love for this to get merged to master and available via pip install :)

[zippolyte]

Thanks for testing this @synth, we'll likely do a release on PyPI this week