Light-the-torch

Cleans up the code and fixes some manual found bugs.

If you want to install this version of `light-the-torch`, use `pip install --pre light-the-torch`.

This is the first step towards supporting newer versions of `pip` in the backend. The major change is that instead of using `pip` internals, `light-the-torch` now actually wraps `pip`.

If you want to install this version of `light-the-torch`, use `pip install --pre light-the-torch`.

Thanks to H00N24 `light-the-torch` now supports PyTorch's LTS releases (39):


$ ltt install --channel=lts torch

This release contains two bug fixes:

1. The precedence of the `AnyBackend`, i.e. a package that can run on any backend since it contains no compiled code, over the other backends lead to installing very old versions in case the library provided non-compiled binaries at first. 37 fixes this.
2. The version comparison to select the best installation candidate was based on string comparison, which leads to false results in case one part of the version crosses the single digit threshold. This was fixed in 38.

33 regressed installation on macOS due to some inconsistencies in PyTorch's naming scheme for earlier versions. 35 fixes this.

This versions adds support for installing any CUDA version that is supported by the nvidia driver rather than only the one that supplies `nvcc`. Shoutout to fepegar for flagging and helping me fixing this!