Parallelize key algorithms for better performance

[goldshtn]

Notably, the graph traversal algorithms can be parallelized (especially the BFS-like ones) by having multiple threads pull from a queue of objects to traverse. Possibly the heap index building can be parallelized as well.

[goldshtn]

Started in 55c016f, there are performance problems around locking heap operations, but the heap doesn't seem to be thread-safe. Filed an issue on microsoft/dotnet-samples#21 for this.

[goldshtn]

Now that I forked ClrMD anyway to make source-level changes, perhaps we can make the memory reader thread-safe, or find some other way to parallelize without having to create multiple DataTargets (which would be incredibly expensive because each one maps the complete dump file to memory).