[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>

docs/design/model_runner_v2.md

@@ -50,7 +50,7 @@ V1 was not originally designed with async scheduling in mind, and support requir
 
 ## 3. Removing Async Barrier
 
-A key requirement for async execution is that CPU operations remain non-blocking. Both explicit sync (for example, `torch.cuda.synchronize`) and implicit sync (for example, unpinned `.to("cuda")`) must be avoided.
+A key requirement for async execution is that CPU operations remain non-blocking. Both explicit sync (for example, `torch.accelerator.synchronize`) and implicit sync (for example, unpinned `.to("cuda")`) must be avoided.
 
 However, async execution can introduce race conditions when CPU and GPU concurrently touch the same memory.
 

docs/usage/troubleshooting.md

@@ -95,7 +95,7 @@ If GPU/CPU communication cannot be established, you can use the following Python
     torch.cuda.set_device(local_rank)
     data = torch.FloatTensor([1,] * 128).to("cuda")
     dist.all_reduce(data, op=dist.ReduceOp.SUM)
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
     value = data.mean().item()
     world_size = dist.get_world_size()
     assert value == world_size, f"Expected {world_size}, got {value}"