Readme update and locally MacOS test (install from repo)

[lexiyutou]

This pull request updates the installation instructions and improves the setup scripts for the PRIMA project, making it easier and more reliable to install and run both from PyPI and from source. The README is clarified for different install scenarios, and the clean_install_local.sh script is enhanced for better Python and PyTorch version handling, as well as more robust dependency installation.

Installation and environment improvements:

• The README now provides clearer, step-by-step instructions for installing PRIMA from both PyPI and source, including explicit Python and CUDA requirements, and improved guidance for macOS users. [1] [2]
• The clean_install_local.sh script now prioritizes Python 3.10, adds more fallback paths for Python executables, and includes Python 3.12 as a last resort. Error messages and usage hints for Python selection are improved.

Dependency management enhancements:

• The script now installs numpy before xtcocotools to ensure build success, installs xtcocotools with --no-build-isolation, and excludes it from the main requirements install to avoid duplicate/conflicting installations.
• Adds logic to detect the CUDA version and select the appropriate PyTorch wheel index URL for installation, or allows override with an environment variable.

Data setup and user experience:

• The demo data download logic is simplified and clarified, with more explicit handling of the --force and --skip-data options.
• The README removes redundant or advanced Hugging Face Space deployment instructions, streamlining the documentation for most users.

Minor documentation tweaks:

• Small formatting and language improvements in the README for clarity and conciseness, such as removing the rocket emoji and updating section headings.

These changes make the installation process more robust, especially for new users and those working on different platforms or with varied CUDA setups.