Fix/softmax out of bounds

[lenarddome]

• Updated the compute methods in both models to handle NaN values in the resulting policies: NaNs are replaced with 1, the policies are renormalized to sum to 1, and a warning is issued to inform the user. [1] [2]
• Added a new test to verify that the Softmax model correctly handles NaN values in activations, ensuring no NaNs remain in the output policies.
• Updated an existing warning test to match the new warning message about NaN handling in policies.

Versioning and documentation:

• Bumped the package version to 0.24.8 in both pyproject.toml and cpm/__version__.py. [1] [2]
• Updated the changelog to document the improved NaN and infinity handling in the decision models.