Drop -XX:+UseParallelGC

[eed3si9n]

Fixes sbt/sbt#5045
Ref sbt/sbt#3559

Currently we set -XX:+UseParallelGC for JDK greater than 9. This isn't a great default because while the peak throughput is somewhat better than the default g1 collector, the worst case performance is really bad with UseParallelGC, especially when the heap size is large.

Given the sudden diversification of JDK implementations, we should stay clear from -XX flags, and let the build users add them if they must.

note

Since there are concerns that we are undoing #189, here's some adhoc local testing.

scala/scala

## g1gc
git clean -fdx && time java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -jar /usr/local/Cellar/sbt/1.3.0/libexec/bin/sbt-launch.jar ";clean;compile;clean;compile;clean;compile"
java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -jar    858.97s user 24.13s system 322% cpu 4:34.00 total

## parallel gc
git clean -fdx && time java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -XX:+UseParallelGC -jar /usr/local/Cellar/sbt/1.3.0/libexec/bin/sbt-launch.jar ";clean;compile;clean;compile;clean;compile"
java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -jar    745.71s user 21.74s system 282% cpu 4:32.06 total

gcg1 is 1% slower.

ThoughtWorksInc/Dsl.scala

## g1gc
git clean -fdx && time java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -jar /usr/local/Cellar/sbt/1.3.0/libexec/bin/sbt-launch.jar ";clean;compile;clean;compile;clean;compile"
java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -jar    439.78s user 20.47s system 555% cpu 1:22.90 total

## parallel gc
git clean -fdx && time java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -XX:+UseParallelGC -jar /usr/local/Cellar/sbt/1.3.0/libexec/bin/sbt-launch.jar ";clean;compile;clean;compile;clean;compile"
java -Dfile.encoding=UTF-8 -Xms2048M -Xmx2048M -Xss2M -jar    441.60s user 20.45s system 550% cpu 1:23.87 total

gcg1 is 1% faster.

[eatkins]

LGTM. We should just be sure to loudly point this out in the release notes.