Update CI environment via PRs in this repository

[rengolin]

We've been having CI failures unrelated to the changes due to environment updates underneath.

We don't want to tie our CI loops to the cluster's own tools and libraries updates, nor we want to hold on the cluster updates due to our own pace in updating them.

But getting random failures is not the way forward. So I propose the following scheme:

1. Tools in the cluster can be updated at will and have no effect on our CI environment
2. Once tools are stable, we add an option to the CI environment to allow us to use it (but not to use it directly).
3. We submit a PR to tpp-mlir to actually use it. This will trigger CI with the new environment and the failures will be linked to an actual PR.

The easiest way to do that is to have tpp-mlir build scripts to set some environment variables, that are picked up by the CI environment setup, which then selects the right tools.

@hfp