GitLab

* Adding USE_HSA_DMA to switch to using hsa_amd_memory_async_copy in lieu of hipMemcpyAsync
* Adding USE_GPU_DMA for A2A benchmark
* Adding largeBAR check and fix for 0-hop GPU-CPU links

* Adding targeted DMA engine support
* Fixing CUDA compilation for H100

* Adding source prep kernel (USE_PREP_KERNEL)
* Adding nvcc-only compilation path
* Fix for NVIDIA - set shared mem usage to 0 by default
* Updating default fill pattern for source data
* Restoring missing example.cfg file