Delete index-cache at the end of container creation

[dbast]

This hopefully is the last attempt to enable libmamba as solver:

• It was initially enabled via Update Miniconda in pkg-build container to py39_23.1.0-1 + enable libmamba solver #350
• Then setting the solver was done after clean to avoid incompatible cache meta data Set solver right after conda clean ... to avoid incompatible caches #355 and Move config reset _before_ all conda config --set calls #356

The container created from current master branch still fails on s390x:

docker run -ti --rm --platform linux/s390x continuumio/anaconda-pkg-build:master /bin/bash -c "conda install numpy" with RuntimeError: Could not read JSON repodata file (/opt/conda/pkgs/cache/8943ae43.json) parse error line 1

But is successful with docker run -ti --rm --platform linux/s390x continuumio/anaconda-pkg-build:master /bin/bash -c "conda clean --index-cache --yes && conda install numpy --yes" ... Thus adding the index-cache clean command at the end of container creation.

[ELundby45]

Thank you!

[dholth]

Since 23.1 conda writes <cache key>.state.json with the http headers, alongside the full, unmodified repodata. It contains a certain set of keys. libmamba requires a different set of keys, and crashes if they are not there; they want to make all keys optional to avoid this problem, but I don't know which release that would be.

We might be able to hack in a conda feature that writes libmamba's expected keys possibly as "" so that this problem does not continually happen.

[dholth]

We checked conda 23.3's .state.json format and it looks much closer to the libmamba format. Hopefully this will work great after the March release.