[che-server] Tune for 32bit JDK.

[jerboaa]

Remove MaxRAMFraction=2 since this would turn out
to be too little heap for a memory limit of 300MB (i.e. 150MB).
Explicitly set MaxHeapSize=256m instead.

Che server VmRSS shows up to be ~240 to ~250MB with 32bit
OpenJDK and other tunings in place. Thus, I've done the exercise
of trying to squeeze che-server into a 300MB limit which seems to
work.

[fabric8cd]

PR now available for testing: https://openshift.io/_profile/_tenant?cheVersion=PR-v26a67a1-1&mavenRepo=https://nexus.cd.test.fabric8.io/content/repositories/staging

[jerboaa]

This PR depends on redhat-developer/rh-che#458 which prepares the che image for this to work. In order to test this with a build I've done use #77.

[fabric8cd]

PR now available for testing: https://openshift.io/_profile/_tenant?cheVersion=PR-v1efb73a-2&mavenRepo=https://nexus.cd.test.fabric8.io/content/repositories/staging

[jerboaa]

Here is a screenshot of this running in my tenant:

This will increase boot-up time of the che pod a little: Ball-park 80secs to 100secs if those numbers printed by Tomcat are anything to go by. Setting -XX:MaxHeapFreeRatio=50 gives roughly the same boot-up time than currently deployed but hurts VmRSS. It's a trade-off to make. Since che-server is mostly long running I've opted for the better long-term memory footprint option.