many core perf fixes

[Maoni0]

This includes a set of many-core fixes, the instrumentation for analyzing heap balancing activities and a GCLogParser tool to parse the logs.

• GetCurrentProcessorNumber returns the correct process number in the CPU group case so we don't always start looking in the first group
• We don't start the initial smoothed gen0 budget as 0 - this causes volatility in the start up stage
• Distribute procs to the right heap/node when there are fewer heaps than procs
• Fixed timing issue when init_numa_node_to_heap_map is called but the proc_no's/numa_node's for heaps aren't fully setup as they are setup from GC threads themselves
• Moved fgn_maxge_percent to per heap as this actuall shows up in remote node access (future work will include making almost all GC's own datastructures like gc_heap numa aware)
• When doing heap balancing we don't look at other heaps less the budget left is small enough. The cost of having to look through all other heaps really shows on HT or with many threads sharing heaps.
• Heap balancing delta is set to be proportional to what's left in the budget instead of a fix ratio of min size.
• Got rid of some unnecessary indirections in heap balancing code.
• Checked in some API changes to facilitate future perf fixes in this area. I had other perf fixes but for 3.0 I'm only checking in these as the other ones require more perf testing. But I want to avoid interface changes.

[Maoni0]

CC @sergiy-k @Andy-MS