Add syncThreads barrier alias

[gshaowei6]

Adds a CUDA-style syncThreads() alias for the existing local work-group barrier. It uses the same Java/JTP barrier path as localBarrier() and emits barrier(CLK_LOCAL_MEM_FENCE) during OpenCL generation.

Tests:

• mvn -s E:\PHD_Project\github\.tools\maven-wikimedia-settings.xml -Dtest=SyncThreadsTest,SyncThreadsSupportTest -DfailIfNoTests=false test
• Also ran the full Maven test suite locally. JUnit reported 579 tests with 0 failures, 0 errors, and 24 skipped; Surefire still marked the run failed because two unrelated forked JVMs (ConvolutionLargeTest, SimpleScalaTest) crashed in native msvcrt.dll on this Windows/Temurin 8 setup.

Fixes #40