fix: unblock GitLab::API::v4 (and Const::Fast) under jcpan

[fglock]

Summary

Investigating jcpan -t GitLab::API::v4 surfaced four independent defects, each masking the next. They all need to land together for the module to load and its unit tests to pass.

1. Parser: use Foo qw() must skip import

Real Perl treats both use Foo () and use Foo qw() as the "skip import" form. PerlOnJava only special-cased () with parens, so use Carp qw(); ran Carp->import() with no args, which Exporter interprets as "import @export defaults" and leaked Carp::croak into the caller. That made sub croak { ... } in GitLab::API::v4::RESTClient look like a redefinition.

Fix: in StatementParser.parseUseDeclaration, also recognise a syntactically empty list expression (e.g. qw()) by walking the parsed AST for ListNode emptiness, gated on the parser having actually consumed tokens. use Foo @runtime_empty_list still calls import(), matching real Perl.

2. HTTP::Tiny stop pretending to export new/get/post/request

Real HTTP::Tiny exports nothing — these are object methods. PerlOnJava's HttpTiny.java added them to @EXPORT, so any package that did use HTTP::Tiny; got Foo::new aliased to HTTP::Tiny::new. Moo's assert_constructor then saw a pre-existing new and bailed with "Unknown constructor for Foo already exists" before installing its own constructor — breaking any Moo class that did use HTTP::Tiny;.

Fix: drop the bogus defineExport("EXPORT", ...). (Pre-existing bug; defect #1 used to make compilation fail earlier and hid it.)

3. Read-only error message format

RuntimeScalarReadOnly.vivify() threw a plain RuntimeException, which made stringifyException() fall through to the multi-line stack-trace formatter:

Modification of a read-only value attempted
        main at FILE line N
        main at FILE line N

Real Perl (and Test::Fatal-style regex matchers in Const::Fast's t/10-basics.t) expect the canonical single-line form:

Modification of a read-only value attempted at FILE line N.

Fix: throw PerlCompilerException instead, so the short-circuit path in stringifyException() runs and the location is appended from PerlCompilerException.buildErrorMessage().

4. Internals::hv_clear_placeholders no-op

Const::Fast calls &Internals::hv_clear_placeholders(\%h) when sealing a hash readonly. PerlOnJava doesn't model Perl's restricted-hash placeholder slots, so there's nothing to clear, but failing with "Undefined subroutine" aborts every call site.

Fix: register it as a no-op stub.

Test plan

• make (full unit-test suite): green.
• ./jcpan -t GitLab::API::v4 now reaches and passes the module's own t/unit.t (Files=3, Tests=2, Result: PASS). Was: fatal compile error.
• Const::Fast t/10-basics.t goes from 5/15 passing (compilation bailed early on hv_clear_placeholders) to 17/25 passing.

The remaining 8 Const::Fast failures are deeper readonly-hash semantics and reassignment-detection (Internals::SvREADONLY returning false on \\do{45} levels, Attempt to reassign a readonly variable not surfacing) — out of scope here, but worth tracking separately.

Generated with Devin