Fix desync between clean and evict

[johnramsden]

There's a problem right now for chunk eviction. In get_evict_targets we remove from the lru and modify the priority queue accordingly.

We then take a subset of what has been modified and clean it after retrieving the relevant zones in get_clean_targets.

The problem is the priority queue can now be potentially adjusted by reads that occur, meaning the priorities can be in an inconsistent state since we do not evict them immediately.

Need to track if chunk is valid, maintain a array of list of chunks representing all zones.

Chunks should be initialized to invalid initially.

When a first write occurs it should be set to valid.

When it's removed from the lru it should be set to invalid and we update the priority queue accordingly.

If we ever do a read, but it is to an invalid chunk, that means we need to update the priority queue.

This also means we do not need to remove from the lru in get_clean_targets, saving us a lot of time. We simply remove the retain, and then in get_evict_targets, when we are removing from the lru we check if it is an invalid entry and if so we remove it and skip it. So basically our lru may end up holding invalid items but that's fine.

[2a46m4]

looks good, just minor question

[johnramsden]

Tested, working