Test exit code should distinguish between failed test and bad args

[tmandry]

Today, if I run a test binary with bad args like --foo-bar, it prints an error message and exits with code 101. Similarly, if I run a test binary correctly but one of the tests fails, it logs the test failure and exits with 101.

It would be nice if we could distinguish between these cases using only the exit code.

Hopefully nothing depends on either of those cases being exactly 101 today, but if we change the code to distinguish between them, we should probably document it and keep it stable from then on.

[fmease]

Triage: No change.