Account for botManagement proxy in structuredClone

[jasnell]

This is ugly and horrible but hopefully temporary. While we are logging the cf.botManagement uses, since we are using a Proxy to catch accesses and Proxy is not supported via structuredClone, we end up having to intercept and handle the case where request.cf is passed off to structuredClone.

It's possible we might have to do this elsewhere if folks are passing request.cf off to other things that using the v8 serializer? I certainly hope not.

[KianNH]

I believe the DO Storage & Queues bindings both use the V8 serializer directly, rather than going through the structuredClone() codepath. i.e:

workerd/src/workerd/api/queue.c++

Lines 14 to 21 in ed865f4

	// Use a specific serialization version to avoid sending messages using a new version before all
	// runtimes at the edge know how to read it.
	jsg::Serializer serializer(isolate, jsg::Serializer::Options {
	.version = 15,
	.omitHeader = false,
	});
	serializer.write(body);
	kj::Array<kj::byte> serialized = serializer.release().data;

workerd/src/workerd/api/actor-state.c++

Lines 726 to 734 in ed865f4

	kj::Array<kj::byte> serializeV8Value(v8::Local<v8::Value> value, v8::Isolate* isolate) {
	jsg::Serializer serializer(isolate, jsg::Serializer::Options {
	.version = 15,
	.omitHeader = false,
	});
	serializer.write(value);
	auto released = serializer.release();
	return kj::mv(released.data);
	}

Would those be covered by this change? My bad if they already are!

[jasnell]

@KianNH ... yeah, good point. Ok, I've updated the PR to account for the additional places. Hopefully this will be temporary and we can pull this back out soon.

[jclee]

Looks OK to me (although I'm not sure I would have necessarily caught the additional serializer uses).

[kentonv]

TBH I would rather just remove the botManagement logging at this point. It seems like it's just not workable. I don't like that this PR adds expense to every serialization operation, and meanwhile it seems very possible that more obscure stuff is broken.

[kentonv]

I suppose this hack only even covers the case where the request.cf object is being serialized. If you place request.cf inside a larger object and serialize that larger object, there is still a problem, I think?