Commit df5c993

These images are tweaked as follows:

**Versions**

* `PYTHON_VERSION`

* `GIT_VERSION`

* `GIT_LFS_VERSION`

* `PANDOC_VERSION`

* `QUARTO_VERSION` (scipy image)

**Miscellaneous**

* `BASE_IMAGE`: Its very base, a [Docker Official Image](https://hub.docker.com/search?q=&type=image&image_filter=official).

* `PARENT_IMAGE`: The image it was built on.

* `BUILD_DATE`: The date it was built (ISO 8601 format).

* `CTAN_REPO`: The CTAN mirror URL. (scipy image)

In addition to the TeX packages used in

[rocker/verse](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_texlive.sh),

[jupyter/scipy-notebook](https://github.com/jupyter/docker-stacks/blob/main/scipy-notebook/Dockerfile)

and required for `nbconvert`, the

[packages requested by the community](https://yihui.org/gh/tinytex/tools/pkgs-yihui.txt)

are installed.

The latest Python version is installed at `/usr/local/bin`, regardless of

whether all packages – such as numba, tensorflow, etc. – are already compatible

with it.

The CUDA and OS versions are selected as follows:

* CUDA: The lastest version that has image flavour `devel` including cuDNN

available.

* OS: The latest version that has TensortRT libraries for both `amd64` and

`arm64` available.

**Versions**

* `CUDA_VERSION`

**Miscellaneous**

* `CUDA_IMAGE`: The CUDA image it is derived from.