Specify version limits for dependencies

[sidtalia]

1. The current pyproject.toml does not specify version limits when installing dependencies.
2. This results in the installation of the latest versions supported on that operating system.
3. This can potentially break the system for someone trying it out
4. As a band-aid, recommend using virtual environments
5. As a fix, update pyproject.toml to use whatever version of torch/numpy is available as long as it is within some bounds (is that possible?)
6. The reason why it did not come up earlier is likely because this was being tested on ubuntu 20.04 (specify this in the README.md).

[sidtalia]

Additionally, restrict default example to use CPU instead of GPU. GPU example require s installing cuda toolkit and breaks if it is not found.