Fix synthetic long local declarations

[KingParmenides]

Summary

Fixes #10.

When Aparapi has to synthesize a local variable table for bytecode compiled without -g, long locals were recorded with descriptor L. In JVM descriptors, L starts an object reference type; scalar long is J. That made the generated OpenCL omit the declaration for locals such as long seed = (long) idx;, producing output like l_3 = (long)i_1;.

Changes

• Map synthetic long local descriptors to JVM descriptor J.
• Add a regression test that compiles a fixture with -g:none and checks that the generated OpenCL declares long l_3 for the cast-initialized local.

Verification

JAVA_HOME=/private/tmp/jdk-11 PATH=/private/tmp/apache-maven-3.9.11/bin:/private/tmp/jdk-11/bin:$PATH mvn -q -Dmaven.repo.local=/private/tmp/aparapi-m2 -DsecondaryCacheDir=/private/tmp/aparapi-zinc-jdk11 -DargLine="-XX:+IgnoreUnrecognizedVMOptions" -Dtest=LongVariableDeclarationWithCastNoDebugInfoTest,FirstAssignInExpressionTest,MultipleAssignTest test
JAVA_HOME=/private/tmp/jdk-11 PATH=/private/tmp/apache-maven-3.9.11/bin:/private/tmp/jdk-11/bin:$PATH mvn -q -Dmaven.repo.local=/private/tmp/aparapi-m2 -DsecondaryCacheDir=/private/tmp/aparapi-zinc-jdk11 -DargLine="-XX:+IgnoreUnrecognizedVMOptions" -DskipTests package

I also ran the full test suite. It reaches 576 tests but fails three existing runtime tests in this environment while repeatedly reporting that the OpenCL native library/device is unavailable: ArrayTest, AccessGetterSetterTest, and CallStaticFromAnonymousKernelTest.

If the bounty is still active and BTC payment is available: 39Q34P8A7g375yqEr8buvNJkUbRgKfbKQZ.