Update rocm xformers dockerfile

[sammysun0711]

What does this PR do?

This PR aim to fix xformers build issue on ROCm platform with composable kernel support.

• Update ROCm version from 6.2.2 to 6.4.4 to support minimal pytorch requirement (torch >=2.7.0)
• Explicit set XFORMERS_CK_FLASH_ATTN=1 and use HIP_ARCHITECTURE instead of HIP_ARCHITECTURES to meet following condition to enable composable kernel during docker build:

  xformers/setup.py

  Lines 576 to 580 in c9911ff

  	elif (
  	torch.version.hip
  	and os.getenv("XFORMERS_CK_FLASH_ATTN", "1") == "1"
  	and (torch.cuda.is_available() or os.getenv("HIP_ARCHITECTURE", "") != "")
  	):

BTW, it seems current setup.py can only support compilation on the single target gfx architecture due to different compilation option:

xformers/setup.py

Lines 606 to 613 in c9911ff

	if arch == "native":
	arch = get_rocm_agent_arch()
	if arch not in ["gfx908", "gfx90a", "gfx942", "gfx950"]:
	raise ValueError(f"Not supported AMD GPU arch: {arch}")
	if arch == "gfx950":
	cc_flag += ["-DFMHA_BUILD_ON_GFX950"]

Before submitting

• Did you have fun?
  • Make sure you had fun coding 🙃
• Did you read the contributor guideline?
• Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
  • N/A
• Did you make sure to update the docs?
  • N/A
• Did you write any new necessary tests?
  • N/A
• Did you update the changelog? (if needed)
  • N/A

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.