fix: preserve int semantics for OpenCL work-item getters

[ItsOtherMauridian]

Summary

• Preserve Java int semantics for OpenCL work-item/size getter mappings by emitting an explicit (int) cast before enclosing widening/arithmetic.
• Exclude getPassId() from that cast because its generated helper already returns int.
• Lower Java long multiplication (LMUL) to an aparapi_lmul helper implemented with 32-bit OpenCL operations, avoiding driver/device miscompilation of native 64-bit multiply.
• Emit the long-multiply helper only when the entrypoint or a called method actually contains LMUL.
• Add regression codegen tests for both relevant cases:
  • (long)getGlobalId() * 100 near Integer.MAX_VALUE to preserve Java getter narrowing semantics.
  • (long)value * 100 to ensure generated OpenCL uses aparapi_lmul(...) instead of native *.
• Update affected codegen snapshots for the explicit get_global_id narrowing cast.

Why this addresses #38

The issue can be hit through two layers:

1. Aparapi maps Java int work-item getters to OpenCL functions that return size_t; without narrowing first, enclosing casts/arithmetic can observe the wrong width.
2. Some OpenCL implementations/devices miscompile native 64-bit integer multiplication; generated kernels should not rely on native long * long for this case.

This PR handles both: it preserves the Java getter return semantics and routes generated LMUL through a helper based on 32-bit operations.

Test Plan

• docker run --rm -v "$PWD":/workspace -w /workspace maven:3.8.8-eclipse-temurin-8 mvn -q -Dtest=com.aparapi.codegen.test.LongMultiplyCastOverflowTest,com.aparapi.codegen.test.LongCastMultiplyTest test
• docker run --rm -v "$PWD":/workspace -w /workspace maven:3.8.8-eclipse-temurin-8 mvn -q -Dtest=com.aparapi.codegen.test.CallGetPassIdTest,com.aparapi.codegen.test.LongCastMultiplyTest,com.aparapi.codegen.test.CompositeArbitraryScopeTest,com.aparapi.codegen.test.ObjectArrayMemberAccessTest,com.aparapi.codegen.test.ObjectArrayMemberCallTest,com.aparapi.codegen.test.ObjectArrayMemberGetterSetterTest,com.aparapi.codegen.test.ObjectArrayMemberHierarchyTest,com.aparapi.codegen.test.ObjectRefCopyTest,com.aparapi.SimpleScalaTest test
• docker run --rm -v "$PWD":/workspace -w /workspace maven:3.8.8-eclipse-temurin-8 mvn -q -Dtest=com.aparapi.codegen.test.*Test,com.aparapi.SimpleScalaTest test

/claim #38

[ItsOtherMauridian]

Small clarification after reviewing the related closed PR #205: this PR now covers both parts of the failure mode for #38.

• It preserves Java int semantics for Aparapi work-item/size getters before widening/arithmetic.
• It avoids native OpenCL 64-bit multiply for Java LMUL by emitting aparapi_lmul(...) backed by 32-bit operations.

I also re-ran the targeted regression checks in Docker:

docker run --rm -v "$PWD":/workspace -w /workspace maven:3.8.8-eclipse-temurin-8 \
  mvn -q -Dtest=com.aparapi.codegen.test.LongMultiplyCastOverflowTest,com.aparapi.codegen.test.LongCastMultiplyTest test