[Device] Adjust threadblock size for gfx950 to increase LL64/Simple performance for AR, RS and AG (#1978)

* Add initial commit to increase tb size to 512
* Fix LL perf issue when subset of NCCL_MAX_NTHREADS is used
Adding a constant to barrier_generic logic from using fallback logic when nthreads < NCCL_MAX_NTHREADS and nthreads == blockDim.X
* Adjust nthreads for LL
* Opt threads for reduce_scatter upper small range
* Add macro for single node
* Restrict MSCCL to 256 threads to prevent mem access fault
* Support pre-MI350 compatibility
* Partially refactor threadblock size override
* Use const macros instead of numerals
* opt out of unused function

[ROCm/rccl commit: 12f51ba8bf]

projects/rccl/src/device/msccl_kernel_impl.h

@@ -28,7 +28,7 @@ extern __shared__ struct mscclShmemData mscclShmem;
 
 inline __device__ static void barrier(int nthreads) {
 #if defined(__HIP_PLATFORM_AMD__) || defined(__HIPCC__)
-  assert(nthreads == NCCL_MAX_NTHREADS);
+  assert(nthreads == MSCCL_MAX_NTHREADS);
   #ifdef __GFX12__
     __asm__ __volatile__("s_waitcnt vmcnt(0) lgkmcnt(0)\ns_barrier_signal -1\ns_barrier_wait -1");
   #else
@@ -96,7 +96,7 @@ __device__ __forceinline__ void mscclRunInterpreter(
   struct ncclDevComm* comm, struct mscclAlgo* algo, struct mscclWork* work) {
   const int tid = threadIdx.x;
   const int bid = blockIdx.x;
-  const int nthreads = NCCL_MAX_NTHREADS;
+  const int nthreads = MSCCL_MAX_NTHREADS;
 
 #if defined(ENABLE_NPKIT)
   uint64_t timestamp_entry = 0;