fatalError in swift_tests executed in RBE produce a trace/BPT trap and no logs right now

If a test fatalErrors, it will kill the whole test runner without any logs. Bazel does correctly report a non-0 exit code at least.

In an RBE like BuildBuddy, this produces almost no useful output.

It would be good to wrap the test runner to handle a test process that exits unexpectedly in order to provide log output and other relevant debugging information from the failed test.

There's been a lot of work on the test runner code in rules_swift lately; I'd be interested in working on this but its not clear how such a change would align to the current work in this area. If anyone can give pointers on a design I'd appreciate the feedback.

Start by reading the test runner code in rules_swift and reproducing a fatalError in swift_tests under an RBE such as BuildBuddy. Done means an unexpectedly exiting test process produces logs and other relevant debugging information instead of only a non-zero exit code.

Written by the indexing model from the issue text.