Remove *_child variants of pre_exec/post_exec hooks

[0xmuon]

Description

Removed all duplicates that I could find.(can you point out other duplicates that I missed? @addisoncrump )
After removing a duplicate implementation, build started failing because run_target was no longer invoking the observers’ pre_exec/post_exec callbacks, which meant things relying on those hooks (stdout/stderr capture observers that finalize reads in post_exec) break. So, I explicitly added the missing pre_exec_all and post_exec_all calls around execute_input_with_command, so observers are always notified at the right time and executor behaves correctly.
fix: #2445

Checklist

• I have run ./scripts/precommit.sh and addressed all comments

[addisoncrump]

Some parts of this are wrong in ways I'm not sure how to handle, because the _child functions are all seemingly misused.

@tokatoka, I see no evidence that we have any _child functions in the whole codebase that do anything. Safe to remove this from the API? It seems to be just incorrectly used everywhere.

[addisoncrump]

This change is not correct; the fuzzer is responsible for invoking pre_ and post_exec on observers. This should be pre_ and post_exec_child, but confusingly, these would not be executed in a child process anyways...

[0xmuon]

oh yeah,will revert to just calling execute_input_with_command since the fuzzer handles observer pre/post.

[addisoncrump]

This change is also not correct; pre_ and post_exec_child is to be executed by the child. However, the original code also appears to be wrong because this code gets executed in the parent. @domenukk / @tokatoka / @andreafioraldi, can you give indication here as to the intended behaviour?

[addisoncrump]

... was this ever correct? I presume this is meant to be pre_exec_child_all, but has the same problem as the other region I highlighted.

[addisoncrump]

Following a conversation with the other LibAFL folks, we've decided to entirely remove the _child variants of pre_ and post_exec.

[0xmuon]

oh sure, I will remove _child variants from observer trait, ObserversTuple, all the override sites in stdio, stacktrace, map, hitcount_map, oom, differential + the inprocess_fork child-side call sites,as of what i see.
Need a lil time to work through it and verify across the affected executors.

[addisoncrump]

Are you sure you want to handle that? I was planning to do so myself, but if you think you've got it, go for it.

[0xmuon]

I will try my best and will ask wherever needed...

[0xmuon]

after the removal, command.rs::tests::test_capture failed: executor.observers.0.output was None.
my take is that StdOutObserver uses an fd-backed capture on linux and macos and only copies bytes into output during post_exec. The old CommandExecutor::run_target was calling pre_exec_all / post_exec_child_all itself, inconsistent with every other executor, where the fuzzer drives observer.So, I removed those internal calls so CommandExecutor matches the rest. test_capture is the only caller that invokes run_target directly without a fuzzer, so nothing was running the hooks anymore.

[0xmuon]

hmm, i don't think the forkserver/libafl-fuzz test fail has occured due to removal of _child observer variant.

thread 'main' panicked at src/main.rs:201:19:
internal error: entered unreachable code

LibAFL/fuzzers/forkserver/libafl-fuzz/src/main.rs

Lines 197 to 204 in b32557c

	let res = run_client(None, mgr, &fuzzer_dir, CoreId(0), &opt, true);
	res
	};
	match res {
	Ok(()) => unreachable!(),
	Err(Error::ShuttingDown) => println!("Fuzzing stopped by user. Good bye."),
	Err(err) => panic!("Failed to run launcher: {err:?}"),
	}

and the test_unicorn_mode (gitlab.com 503 during AFL++ submodule clone) failed due to gitly outage