Fix leak checking

[pimyn-girgis]

This PR enables the periodic execution of a leak checker within the executor, it also fixes the parsing of kmemleak output.
Fixes #4728.

[a-nogikh]

An alternative implementation would be to e.g. add some new message type to trigger leak checking here:

syzkaller/executor/executor_runner.h

Lines 612 to 621 in 65a0eec

	if (auto* msg = raw.msg.AsExecRequest())
	Handle(*msg);
	else if (auto* msg = raw.msg.AsSignalUpdate())
	Handle(*msg);
	else if (auto* msg = raw.msg.AsCorpusTriaged())
	Handle(*msg);
	else if (auto* msg = raw.msg.AsStateRequest())
	Handle(*msg);
	else
	failmsg("unknown host message type", "type=%d", static_cast<int>(raw.msg.type));

And then, in https://github.com/google/syzkaller/blob/master/pkg/rpcserver/runner.go, every X requests/Y minutes make sure there are no pending requests and start the leak checking process. Once leak checking is finished, continue fuzzing.

I am hesitating re. what'd be better here - the C++ implementation is shorter, but makes the poorly tested logic even more complicated, while the (mostly) Go implementation would be bigger than the 50 line change we have here.

[a-nogikh]

Otherwise looks good to me

[a-nogikh]

I think it should all work fine, but let's double-check it before merging. Please run the modified syzkaller for at least several hours on

1. A Linux kernel without leak checking
2. A Linux kernel with kmemleak enabled

(You can e.g. set it up overnight)

We should ensure that in both cases there are no obvious syzkaller bugs, that fuzzing progresses and reproducers are found.
I am also curious about the difference in exec/sec between the two cases.