Notes on GC code

"The GC_heap structure now defines the global nursery, while each processor's GC_state tracks the per-processor allocation area. Allocation from a processor-local area occurs just as for the single- processor version of MLton. When processor-local space is exhausted, the processor attempts to take a chunk from the global space using an atomic compare-and-swap operation. If no global space is available, the processor signals that a collection is necessary and waits for the remaining processors to catch up."