[Nestmates] RFR: 8197402: [Nestmates] Update javac tests that assume existence of synthetic nested classes for constructor access

[David Holmes]

webrev: http://cr.openjdk.java.net/~dholmes/8197402/webrev/
bug: https://bugs.openjdk.java.net/browse/JDK-8197402

This fix is contributed by Maurizio. I've tested it locally and in our 
tier1-3 platform testing.

 From Maurizio:

Tests are too dependent on the javac output and they strictly verify 
that the (synthetic) inner class corresponding to a private constructor 
token is emitted; with nestmates such token classes are not needed, so 
they are not in the output. The tests need to change.

I've tried, as much as possible, to extend the tests so that they run 
both in pre-nestmate mode and in nestmate mode, so that we check all 
combinations.

One test (NoAbortOnBadClassfile) could not be adjusted to work on 
nestmate mode (the test is too fragile and relies on the exact contents 
of the inner classes attributes, which are slightly changes because of 
the new class references available in the nestmate attributes). So for 
now that is only ran in pre-nestmate mode; Jan is looking at ways to 
make the test more reliable, but we're ok with going with this patch for 
now and maybe fix this test later.
---

These changes all seem fine to me.

Thanks,
David