Add tutorial: qlearning with/without action masking for Taxi v3 env

[dantp-ai]

Description

Adds a new tutorial demonstrating difference in learning with and without action masking for Taxi v3 environment using Q-learning.

Fixes #28 (Task: "How to use the action sample masking, with example from Taxi")

Type of change

• Add new tutorial

Checklist:

• I have run the pre-commit checks with pre-commit run --all-files (see CONTRIBUTING.md instructions to set it up)
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings

[pseudo-rnd-thoughts]

Thanks for the tutorial @dantp-ai, Could you update the project main as the Test Gymnasium Tutorial CI was broken and is now fixed

[dantp-ai]

The Test Gymnasium Tutorial job failed.

I don't understand the output from this job:

Run if [ "" -eq 0 ]; then
/home/runner/work/_temp/618a3580-acb8-4abc-aec3-99c[21](https://github.com/Farama-Foundation/Gymnasium/actions/runs/14179527178/job/39722202860?pr=1345#step:11:22)ed1f93a.sh: line 1: [: : integer expression expected
Notice: All  tutorials passed.
/home/runner/work/_temp/618a3580-acb8-4abc-aec3-99c21ed1f93a.sh: line 3: [: : integer expression expected
/home/runner/work/_temp/618a3580-acb8-4abc-aec3-99c21ed1f93a.sh: line 11: [: : integer expression expected
Error: Process completed with exit code 2.

Seems to be related to line 11 from the run-tutorial.yml but I don't understand how that can fail?

[dantp-ai]

P.S.: I merged latest main into my branch.

[pseudo-rnd-thoughts]

Thanks @dantp-ai for updating, there was a bug in the testing that should be fixed now if you merge main again.

On the tutorial, the training looks good, but I would love to see a larger discussion about what an action mask is (how to get the data and space.sample) and why it can improve performance.

Could you center the tutorial on what action masking is, how to use it, and why it can improve performance with helpful code examples for each. Before moving into the training section as in reality, there is only 2 lines differences to "normal" training if that makes sense.

Finally, I would remove the run_experiment function and just merge the functionality into the other function.

If you make those changes, then I think we should be good to merge