Rerun fails with coverage and "dependent" subtests

I've started to use the `--rerun-fails` option and noticed that when coverage profiling is enabled, the produced `coverage.out` file only contains the coverage from the failed but re-executed sub-tests.

For example:
```
$ gotestsum --rerun-fails --packages=. -- -coverprofile=coverage.out -coverpkg="." && go tool cover -func coverage.out
✖ . (2ms) (coverage: 76.9% of statements in .)

DONE 4 tests, 2 failures in 0.271s

✓ . (2ms) (coverage: 46.2% of statements in .)

=== Failed
=== FAIL: . Test_sayHello/Voldemort_makes_it_fail_every_second_time (0.00s)
main_test.go:16:
Error Trace: /home/agr/oss/gotestsum-rerun-dependent-subtests/main_test.go:16
Error: Received unexpected error:
afraid of Voldemort, cannot say hello
Test: Test_sayHello/Voldemort_makes_it_fail_every_second_time
--- FAIL: Test_sayHello/Voldemort_makes_it_fail_every_second_time (0.00s)

=== FAIL: . Test_sayHello (0.00s)

DONE 2 runs, 6 tests, 2 failures in 0.509s
github.com/neiser/gotestsum-rerun-dependent-subtests/main.go:11: sayHello 46.2%
total: (statements) 46.2%
```

The above example usage is from [this little demo project to illustrate my issue](https://github.com/neiser/gotestsum-rerun-dependent-subtests), see `run-test.sh` script there.

I've tried to catch the `coverage.out` from the first run using with `--post-run-command="bash -c 'mv coverage.out $(mktemp coverage.out.XXXXXXXXXX)'"`, but that only produced one file. I assume that's because the post-run command is only executed after all re-runs are done by gotestsum.

Do you have a suggestion how to calculate coverage correctly and/or generate a "merged" coverage.out file from all re-runs?

Futhermore, I'd like to see a command line option such as `--rerun-all-subtests` which re-runs not only one specific failed sub-test, but essentially all tests of the test case containing the failed sub-test (i.e. find the root of a failed subtest separated by `/` and run that one again). That would help when a sophisticated integration test is split up into sub-tests and subsequent sub-tests need all previous sub-tests to be run to be successful. I can also prepare an illustration to this problem, but I hope you get the idea. What do you think?

Reproduce the behavior using the command and run-test.sh script from the linked demo, with the failing case shown in main_test.go. Start by tracing how gotestsum runs failed subtests and invokes --post-run-command. Done means coverage from the initial run and reruns can be combined, and the requested behavior for rerunning all subtests of a failed test case is defined and tested.

Written by the indexing model from the issue text.