RFR: 8251354: Shenandoah: Fix jdk/jfr/tool/TestPrintJSON.java test failure

[Roman Kennke]

Please review the following issue and fix. This is a 11u-only and
Shenandoah-only bug, but affects shared code.

With the recent Shenandoah backport to jdk11u, we decided to optionally
split metadata.xml so that Shenandoah can provide its own metadata-
shenandoah.xml in a separate file. This avoids generating code for
Shenandoah when Shenandoah is not present in the build.

However, the metadata.xml is also read at run-time, but metadata-
shenandoah.xml is not present at run-time, nor would it be read with
the current implementation of MetadataHandler.java. This makes the test
jdk/jfr/tool/TestPrintJSON.java fail.

I propose to revert the parts of the Shenandoah patch that allows
breaking-up metadata.xml. The downside is that it spills a tiny amount
of (empty) Shenandoah-related code in builds without Shenandoah.
However, it is the exact same thing that also happens for any other GC.
On the positive side, this seems the most conservative approach because
that is what has been tested for a very long time before we integrated
Shenandoah into 11u.

Alternatively, we could further extend the makefile machinery and the
MetadataHandler.java to also (optionally) read metadata-shenandoah.xml
at run-time. It would have the advantage of cleanliness in non-
Shenandoah builds, and the disadvangate of further divergence from
upstream jdk16, and risk of rather untested code, and uncertainty
whether or not this would be the last such issue.

Bug:https://bugs.openjdk.java.net/browse/JDK-8251354
Webrev: http://cr.openjdk.java.net/~rkennke/JDK-8251354/webrev.00/
Testing: build with and without Shenandoah, JFR tests with and without
Shenandoah

What do you think?

Roman