We are currently doing this

  pflag.Usage = func() {
    r.RunUsageExit()
  }

That usage gets invoked if pflag fails to parse (e.g. due to an unknown
flag), but then rather than letting pflag exit with a non-0 error code,
the RunUsageExit() func exits with a 0 code, misleading callers into
thinking we had a successful script invocation.

That's a bug. We seem to be able to safely just remove the 'exit' part
to fix it and still have everything else working properly.

I took this opportunity to investigate the 'double unknown error' print
issue a bit. It's a bug in the pflag lib, I'm not the only one to have
found it: spf13/pflag#352

Unfortunately pflag is not maintained anymore (last release >5 years
ago), so very small chance this is ever getting fixed.

The only solution (bar forking pflag, but I don't think this is a
legitimate reason to) is to switch to using our own flagset so that we
can control the error handling to be ContinueOnError instead of
ExitOnError. This way, we can then avoid the double print and handle the
error ourselves. Anyway, that's a future ticket.

We moved to using our own flagset in the previous commit, which lets us
then instantiate it with 'ContinueOnError', and catch the parsing error
ourselves and print it once, rather than being stuck with the pflag bug
that results in the error being double printed
(spf13/pflag#352).