RFR: Have adaptive heuristic include only full cycles in average cycle time

[William Kemper]

We have observed cases when a high percentage of short cut cycles (i.e., when evacuation and update reference phases are skipped) may cause the heuristic to believe it can complete _any_ cycle in the time it takes to complete an abbreviated cycle. Triggering the cycle too late may lead to a degenerated cycle. Abbreviated "learning cycles" are similarly ignored. 

A command line option has been added to disable this change in behavior and restore the previous behavior of including every cycle's time in the average time.

-------------

Commit messages:
 - Add feature flag to re-enable inclusion of all cycle times in heuristic
 - Only include full cycles in heuristic's average time

Changes: https://git.openjdk.java.net/shenandoah/pull/139/files
 Webrev: https://webrevs.openjdk.java.net/?repo=shenandoah&pr=139&range=00
  Stats: 29 lines in 10 files changed: 15 ins; 0 del; 14 mod
  Patch: https://git.openjdk.java.net/shenandoah/pull/139.diff
  Fetch: git fetch https://git.openjdk.java.net/shenandoah pull/139/head:pull/139

PR: https://git.openjdk.java.net/shenandoah/pull/139