Better suport custom & plugin algos

[nicolasservel]

See ch59584

This PR:

• adds better support for plugin & custom algorithms in the ML helpers, in particular when listing & getting algorithm settings
• adds support for wait_guess_complete when creating clustering ML Task, to align with prediction MLTasks for which it is already available, and changes the default behavior to wait_guess_complete=True, which is what we want in the vast majority of cases
• fixes a few things found while testing:
  • no fail if "plugin_python" key not in modeling, for clustering task when disable_all_algorithms
  • only register "impurity" for MLlib Random Forest, duplication previously broke instantiation of the algorithm settings

[shortcut-integration]

This pull request has been linked to Clubhouse Story #59584: Better support for custom & plugin algos in public ML api helpers.

[mhham]

Minor remarks

[samronsin]

Just need to fix the _get_plugin_algorithm_names method which breaks get_all_possible_algorithm_names and its downstream methods with Python 3.

[samronsin]

LGTM !