Slow initialization for VM map

[qinsoon]

Following our discussion in #632 (comment), after fixing the initialisation cost for SFT map in #632, our boot time is still significantly slower than the stock OpenJDK.

See the table below (from #632 (comment)):

	Elapsed wall clock time (s)	Maximum resident set size (kbytes)	thread attach log time
Stock OpenJDK	0.17	60696	0.021s
MMTk - sparse chunk map (master)	0.52	839596	0.349s
...	...	...	...
MMTk - space map (PR)	0.30	314604	0.151s

perf suggests the time is spent in initializing the MMTk instance (MMTK::new()). I instrumented the method, and found that 93% of the time in MMTK::new() is spent in VMMap::new().

Our Map64 is a straightforward port of https://github.com/JikesRVM/JikesRVM/blob/master/MMTk/src/org/mmtk/utility/heap/layout/Map64.java from the Java MMTk. I am not familiar with the code, and I haven't looked further into the problem.