Never trust output to be UTF8, always ignore errors

We just had another case of test job crashing because the output was not UTF8.

I modified all the "decode to UTF8" calls I could find to ignore problems.