fix: use cache sharing=locked,id=${TARGETPLATFORM}
We're continuing to see incorrect execv target platform architectures being used. This continues to push through the separation of caches based on target platform as id and using locked access to avoid having to rely on the internal locking mechanisms.