Add Wheel & Docs: Flash Attention 2.7.4.post1 for Py3.12 / CUDA 12.1 / PyTorch 2.5.1

This PR adds support for Python 3.12 users by providing a pre-compiled wheel for flash-attn v2.7.4.post1 and updating the documentation accordingly.

Wheel Details:

• Package: flash-attn
• Version: 2.7.4.post1
• Target Environment:
  • Python: 3.12.x (cp312)
  • CUDA Toolkit: 12.1
  • PyTorch: 2.5.1+cu121
  • OS: Windows x64
• Build Context: Successfully built using Visual Studio 2022 LTSC v17.4.x toolchain.

Wheel File Location:

• The .whl file (flash_attn-2.7.4.post1-cp312-cp312-win_amd64.whl) has been uploaded as a binary asset to the release tagged v2.7.4.post1_py312_cu121_torch251 on my fork (Wisdawn/flash-attention-windows). It is not included directly in this PR's file changes due to GitHub's file size limits.

README.md Updates in this PR:

• Adds the Py3.12 wheel details to the "Available Wheels" section.
• Adds prerequisites for the Py3.12 wheel under "Requirements".
• Updates the "Installation" section to direct users to download wheels from the Releases page.
• Adds SHA256 checksum for the new wheel under "Security".
• Adds a detailed example build process for Py3.12/CUDA 12.1 under "Instructions for Building New Wheels".
• Adds specific troubleshooting steps relevant to the Py3.12/CUDA 12.1 build under "Troubleshooting".
• Adds a "Contributing Wheels" section explaining the PR process.
• Minor updates to "Known Issues" and the main title.

Hoping this contribution helps other Windows users! Let me know if any changes are needed.