fix: suppress encoding errors during testing

In CI the test script thinks the system locale is ASCII, but apparently one of
the subprocesses gives us non-ASCII data. Rather than trying to deal with this
mess, just ignore encoding errors when piping data back to stdout/stderr.