fix(values,GCString): values %h scalar context, split GCString shim into own file

[fglock]

Summary

Two fixes uncovered while investigating jcpan -t User::Identity. The full dependency chain is User::Identity → Log::Report → Log::Report::Util → String::Print → Unicode::GCString, and it was failing at every step.

1. values %h in scalar context as a sub's last expression

my %h = (a => "X", b => "Y", c => "Z");
sub v { values %h }
my $n = v();   # was: "Z"   now: 3

RuntimeHash.keys() set scalarContextSize on the resulting RuntimeArray so getList() wrapped (rather than copied) it; values() did not, so the resulting list scalarized to its last element instead of the count. Affected both JVM and interpreter backends.

Fix: mirror keys() and set list.scalarContextSize = list.elements.size() in both the plain and TIED_HASH paths of RuntimeHash.values().

This was the root cause of t/30col.t failures in User::Identity, where User::Identity::Collection::roles is defined as sub roles() { values %{ $_[0]->{UIC_roles}} } and called via cmp_ok($col->roles, '==', 1, ...).

2. Unicode::GCString shim wasn't a discrete file

The pure-Perl shim of Unicode::GCString lived as a second package inside lib/Unicode/LineBreak.pm. Two problems followed:

• use Unicode::GCString (without first loading LineBreak) couldn't find a .pm file, so it failed with "Can't locate Unicode/GCString.pm". String::Print does exactly this.
• MakeMaker's "skip files already bundled in the PerlOnJava JAR" logic only saw Unicode/LineBreak.pm. Installing CPAN's Unicode-LineBreak distribution would therefore install/shadow the shim with the XS-needing Unicode/GCString.pm from blib, which then died with "Can't locate object method _new".

Fix: move the shim into its own file src/main/perl/lib/Unicode/GCString.pm and have LineBreak.pm require it. Now both files are visible under jar:PERL5LIB and the SKIP logic preserves both:

SKIP: Unicode/GCString.pm (bundled in PerlOnJava JAR)
SKIP: Unicode/LineBreak.pm (bundled in PerlOnJava JAR)

Result

• jcpan -t User::Identity now passes all 114 subtests across 3 test files (was 4/4 subtests failing because use User::Identity blew up immediately).
• String::Print test pass rate jumps from 1/19 test files to 15/19; remaining failures are wide-character column-width formatting in the shim, unrelated to this PR.

Test plan

• make (full unit test suite) — passes
• ./jperl -e '...' and ./jperl --interpreter -e '...' — both yield count 3 for values %h in sub-scalar context
• ./jcpan -t User::Identity — Result: PASS, all 114 subtests successful

Generated with Devin