RFR (S): [8u] Re-implement verbose GC logging

[Aleksey Shipilev]

Hi,

This makes Shenandoah print saner -verbose:gc in 8u, matching what 9 prints:
  http://cr.openjdk.java.net/~shade/shenandoah/jdk9-8u/verbose-gc-logging/webrev.01/

Ended up doing our own GCTraceTime implementation. Also touches up handling of
log(gc,ergo) printouts that are disabled by default on 9. In 8u current macros
for log(...) spam into stdout.

Testing: hotspot_gc_shenandoah

Output on 9:

[15.015s][info][gc] GC(1) Pause Init Mark 8.219ms
[15.406s][info][gc] GC(1) Concurrent marking 5115M->5783M(10240M) 390.467ms
[15.410s][info][gc] GC(1) Pause Final Mark 5783M->4823M(10240M) 3.562ms
[15.451s][info][gc] GC(1) Concurrent evacuation  4823M->5287M(10240M) 40.576ms
[15.451s][info][gc] GC(1) Concurrent reset bitmaps 0.089ms

Output on current 8u:

predicted cset threshold: 28
Starting concurrent mark at 2322432K CSet (22%)
24.076: [Pause Init-Mark, 0.0056708 secs]
24.711: [Pause Final Mark, 0.0067279 secs]

Output on patched 8u:

31.314: #4: [Pause Init Mark, 0.967 ms]
31.315: #4: [Concurrent marking 8964M->9690M(10G), 619.847 ms]
31.935: #4: [Pause Final Mark 9690M->6725M(10G), 9.763 ms]
31.945: #4: [Concurrent evacuation 6729M->7163M(10G), 52.714 ms]
31.998: #4: [Concurrent reset bitmaps, 3.247 ms]


Thanks,
-Aleksey