RFR: Enable TLHS by default (again)

[Roman Kennke]

>> Do we want to finally enable TLHS for good? Or do we considere too risky
>> compared to (almost zero immediate) usefulness? It'd allow some more
>> concurrent processing going forward.
> 
> I think it is too risky without apparent benefit, at this point. Adding the simple regression test
> with TLHS specifically enabled would be nice, though.

How about now? Upstream is deprecating the flag, and working towards
removal already: https://bugs.openjdk.java.net/browse/JDK-8220049

The single program that I remember having problems with is the Serial
benchmark with +TLHS and +traversal, and I've just run it 30 times in a
row without issues. I've also run our regression tests, all of specjvm
and specjbb with no issues. I see indications that this might actually
help throughput and latency, but this probably needs more testing. It
doesn't look like it would regress throughput or latency though.

I propose to push it to sh/jdk now, let it test+bake for a bit, then
upstream it to jdk/jdk. We should probably not backport to jdk11. Not sure.

WDYT?

Roman