JDK-8223051: support loops with long (64b) trip counts

[John Rose]

FTR:  I posted more thoughts about generalizing our C2 loop transforms from int to long here:

  https://bugs.openjdk.java.net/browse/JDK-8223051

This is not urgent, but will become more and more important as we adapt our APIs to today’s 64-bit realities.

One suggestion:  Rather than duplicate all of our int-specific loop transform logic to long-specific logic,
keep just the old copies of everything but upgrade them in place to handle intptr_t (or just long, frankly).

One way to test the intptr_t tactic is to put the logic on a dynamic switch (int vs. long tripcounts) and
use A/B testing to shake out bugs on our workloads.  For normal use, set the switch (by default) to sense
the running platform (ILP32 vs. LP64).

That might be over-clever; perhaps we just want to generalize all the loop transform logic.  But even so,
the intptr_t idea (with a switch) could provide a temporary testbed for the above sort of A/B testing.

— John