Negative slop accounting with the non-moving gc

I've noticed that occasionally GHC's slop metric underflows when using the non-moving GC. As far as I can tell this happens during the concurrent mark phase due to some incorrect accounting. I haven't investigated this properly yet, but I'm making an issue so I don't lose track of this.

Part of this should be adding an assertion to make sure that slop isn't underflowing to make this easier to catch in the future.