fuzzer cannot terminate gracefully

[mlugg]

Split from #24588.

When using zig build --fuzz, if the "Rebuild" button is pressed in the web UI, you will see a panic with this message:

TODO: terminate the fuzzer processes

This should be fairly straightforward to fix; I just didn't want that branch to balloon any further. There are fuzzWorkerRun tasks running which need to somehow gracefully terminate their children. Graceful termination is important to avoid the serialized fuzzer state from being left in an incorrect state.

[andrewrk]

Note that graceful termination of fuzzing processes is decidedly not important, by design. When a fuzzing process finds a bug, it may crash (such as reaching a panic or segmentation fault), or may be terminated due to timeout. Therefore, the system is designed such that the supervisor process handles such crashes as part of the normal operation.

Related:

• introduce --fuzz=<limit> to make tests pass after a certain quantity of fuzz testing has completed #22900
• fuzzing infinite mode #22901

[mlugg]

It's true that it's okay for the child to crash during test execution, but killing the child at a random point is problematic, because it could be killed partway through writing some data to the mapped coverage file, leaving that file in an incorrect state. That said, I'm somewhat dubious of the mmap strategy anyway -- see #24656 -- so if that were to change, it could be fine to kill the child at any point.