feat(Kernel): add local message passing helpers

[Silas-Sitri]

Summary

• add Kernel.localExchange(...) for integer work-item value exchange inside a work-group
• add Kernel.localBroadcast(...) for integer work-group broadcasts
• emit OpenCL helper functions only when the corresponding message-passing API is used
• add codegen and Java thread-pool runtime coverage

Notes

These helpers use Aparapi local memory and local barriers. Callers must pass a @Local buffer, or a buffer named with Kernel.LOCAL_SUFFIX, and all work-items in the work-group must reach the calls in the same order.

Closes #43

Testing

• /tmp/apache-maven-3.9.16/bin/mvn -q test-compile
• JAVA_HOME=/usr/lib/jvm/java-17-openjdk-amd64 /tmp/apache-maven-3.9.16/bin/mvn -q -DskipTests package
• manual JUnit: java -cp ... org.junit.runner.JUnitCore com.aparapi.codegen.test.LocalMessagePassingTest com.aparapi.runtime.LocalMessagePassingTest -> OK (3 tests)
• git diff --check

Full mvn test could not run in this local Java 17 environment because the existing Surefire config passes the Java 8-era option -XX:MaxPermSize=512m, which Java 17 rejects before tests start.