better support for from-OS large heap and cache allocations

I was experimenting with using large allocations from the OS instead of DR's -vm_reserve and hit several issues with the current runtime options which I plan to fix:

• allow -no_vm_reserve to work Today get an assert: core/heap.c:1034 vmcode_get_start() >= heap_allowable_region_start

• don't make me set trace unit sizes to be bigger than -cache_commit_increment when I have -disable_traces

• let me raise -heap_commit_increment larger than signal special heap unit size w/o having to raise -max_pending_signals to some ridiculous amount