Fork detection: add rfork() and similar support for FreeBSD/OpenBSD.

On BSD systems, `pthread_atfork()` is insufficient, as e.g. `rfork()`
(similar to Linux's `clone()`) can create processes too, and will not
call pthread's handlers. Use BSD's `minherit` mechanism to detect this
kind of process duplication the same way as it works on Linux.

There would be similar problems if one were to use `vfork()` on some
systems. However, do note that according to POSIX.1, doing virtually
anything in the `vfork()` child is undefined behavior, and we can't
reasonably be expected to guard against undefined behavior in callers.
macOS even deprecated the function and shows a warning for any use of
it.

I believe this will not actually hit any real-world software, and am
merely adding this to be sure.

Manually tested on OpenBSD:

```
grawp$ uname -a
OpenBSD grawp.my.domain 7.1 GENERIC.MP#465 amd64
grawp$ ./crypto_test --gtest_filter='*Fork*'
Note: Google Test filter = *Fork*
[==========] Running 3 tests from 2 test suites.
[----------] Global test environment set-up.
[----------] 2 tests from ForkDetect
[ RUN      ] ForkDetect.Test
[       OK ] ForkDetect.Test (7 ms)
[ RUN      ] ForkDetect.TestAlternate
[       OK ] ForkDetect.TestAlternate (0 ms)
[----------] 2 tests from ForkDetect (8 ms total)

[----------] 1 test from RandTest
[ RUN      ] RandTest.Fork
[       OK ] RandTest.Fork (4 ms)
[----------] 1 test from RandTest (4 ms total)

[----------] Global test environment tear-down
[==========] 3 tests from 2 test suites ran. (13 ms total)
[  PASSED  ] 3 tests.
```

and on FreeBSD:

```
[rpolzer@grawp build (git)-[d03c3135d]-]$ uname -a
FreeBSD grawp 15.0-RELEASE-p6 FreeBSD 15.0-RELEASE-p6 releng/15.0-n281024-6b6bc9afa0b0 GENERIC amd64
[rpolzer@grawp build (git)-[d03c3135d]-]$ ./crypto_test --gtest_filter='*Fork*'
Note: Google Test filter = *Fork*
[==========] Running 3 tests from 2 test suites.
[----------] Global test environment set-up.
[----------] 2 tests from ForkDetect
[ RUN      ] ForkDetect.Test
[       OK ] ForkDetect.Test (2 ms)
[ RUN      ] ForkDetect.TestAlternate
[       OK ] ForkDetect.TestAlternate (0 ms)
[----------] 2 tests from ForkDetect (3 ms total)

[----------] 1 test from RandTest
[ RUN      ] RandTest.Fork
[       OK ] RandTest.Fork (1 ms)
[----------] 1 test from RandTest (1 ms total)

[----------] Global test environment tear-down
[==========] 3 tests from 2 test suites ran. (5 ms total)
[  PASSED  ] 3 tests.
```

Tested on both that without this change, the added test fails.

Update-Note: if you seriously sidestep the OS-level proper channels of
creating a new process duplicating address space, stop doing that, or at
least do not call BoringSSL from the so-forked process. In general, any
address space duplication shall be done using `fork()` and `fork()`
only.

Change-Id: I11f0c3e868a36ca5a1ab8b605c40f9856a6a6964
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/94247
Commit-Queue: Rudolf Polzer <rpolzer@google.com>
Presubmit-BoringSSL-Verified: boringssl-scoped@luci-project-accounts.iam.gserviceaccount.com <boringssl-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: David Benjamin <davidben@google.com>