MPL.GC structure for mpl-specific gc statistics

[shwestrick]

This makes some progress on #103 and #106.

The MPL.GC structures includes some statistics, including:

• number of local GCs
• time spent in local GC (wall-clock)
• time spent in promotion before local GC (wall-clock)
• bytes allocated (currently a bit funky, because includes fragmentation due to block allocation strategy)
• bytes reclaimed by local GC (also funky, due to including benefits of defragmentation)

All of these statistics come in both global and per-processor flavors. The global amount is just the sum of per-processor values; it's just for convenience.

There is also a MPL.GC.currentHeapSize, but this not implemented yet. I plan to implement this in the future.