CI fix and Deep CFR refactor

[alexunderch]

Brought back commented workflows just by replacing CIBW_MANYLINUX image from manylinux_2014 that's hit its EOL by:

 CIBW_MANYLINUX_X86_64_IMAGE: manylinux_2_28
 CIBW_MANYLINUX_AARCH64_IMAGE: manylinux_2_28

Added handselected pinned requirements.

[alexunderch]

@lanctot I tried to unsuccessfully separate the development chain...
Proof that this runs:

• https://github.com/alexunderch/open_spiel/actions/runs/20965202386/job/60253507355
• https://github.com/alexunderch/open_spiel/actions/runs/20964252894

[lanctot]

@lanctot I tried to unsuccessfully separate the development chain... Proof that this runs:

• https://github.com/alexunderch/open_spiel/actions/runs/20965202386/job/60253507355
• https://github.com/alexunderch/open_spiel/actions/runs/20964252894

No problem. If tests pass... how about we just use this PR both as a fix for #1441 and as a replacement for #1408 ?

[alexunderch]

as you wish!

[lanctot]

Can you try re-enabling the Linux arm64 tests too ?

[alexunderch]

@lanctot aren't they enabled?..

[lanctot]

Can you do one more commit to re-enable both of these commented Linux tests, in this PR, so they get triggered?

[lanctot]

@lanctot aren't they enabled?..

No probably because of the merge conflicts. See screenshot above: the "Files Changed".. wheels.yml shows them now as disabled again because they're currently commented out in master

[alexunderch]

Hmmm, on my branch they are active, let me look once more https://github.com/alexunderch/open_spiel/blob/39c250a3f53eb2407ec17467524de9f546718c20/.github/workflows/wheels.yml#L43

[lanctot]

Weird. I tried something (invoking my power as admin) and was able to make the commit even though it was in your branch. Cool. Now showing up in Flies Changed; let's see what happens!

[alexunderch]

I think me and my interesting branching habits broke something. Will do my best not to happen in the future

[alexunderch]

Sorry, I forgot to re-add the test for deep_cfr. But it's a good ending. Just another check and we're good to go. It was tough, but we did it, only sacrificing python3.10 and MacOS build for Intel.

[lanctot]

Sorry, I forgot to re-add the test for deep_cfr. But it's a good ending. Just another check and we're good to go. It was tough, but we did it, only sacrificing python3.10 and MacOS build for Intel.

No problem. And: totally! Thanks for all your work over the past few days!

[alexunderch]

It's only the first out of several big PRs. However, I believe that the next ones will be much simpler, because we would only need to focus on the tests rather than all of these dependencies...

[lanctot]