Enhance Packager to make the MAP more secure, easier to use, and based on new version of PyTorch image for newer CUDA versions.

[MMelQin]

This PR intends to improve the packager itself, as well as making the MAP based on newer if not the latest PyTorch image, by default, to make use of newer version of CUDA.

• Packager creates the env var in the MAP image, with the default value being the path specified in the spec. This will eliminate the need to provide the env var when docker run. But the MONAI_MODELPATH was missing. Hence added
• Resolve a minor security risks where /tmp is used for temp file in the Packager
• The default PyTorch image is now nvcr.io/nvidia/pytorch:22.08-py3, which is the last version that still supports Nvidia Driver release 515 (or later). Also, this base has PyTorch 1.13
• The minimal version of NumPy is bumped to 1.21.6, the latest version still compatible with Python 3.7, which is still supported by App SDK.
• Fixed a imshow issue in the simple_app Jupyter Notebook
• Tested all example apps and notebooks, both local execution as well as in MAP

Signed-off-by: M Q mingmelvinq@nvidia.com

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

No Coverage information
0.0% Duplication