Codecaching: support saving code compiled across multiple threads in SaveData

Locally I have savedata taking a span of ThreadState and then copying them all into a common vector but not sure I like that too much.
I was thinking it would be nicer if the frontend had access to TakeRelocations or a similar API and it could pass in a full relocations vector, or if there was a global relocations pool in CTX that we copied to after compilation. (I don't think contention of that would be much worse than what we have already)