Can't initialize multiple classifiers in the same python session

[jluethi]

If a user closes the current classifier widget window and initializes a fresh classifier, that doesn't work. Multiple issues:

• What about the pre-existing selection & prediction layers?
• Keybinding issue: Given that I'm specifying key binding when launching the classifier widget, this conflicts with re-inializing and throws this error:

  
  File ~/opt/miniconda3/envs/napari-classifier/lib/python3.9/site-packages/napari/utils/key_bindings.py:277, in bind_key(keymap={'0': <function ClassifierWidget.create_selector_widget.<locals>.set_class_0>, '1': <function ClassifierWidget.create_selector_widget.<locals>.set_class_1>, '2': <function ClassifierWidget.create_selector_widget.<locals>.set_class_2>, '3': <function ClassifierWidget.create_selector_widget.<locals>.set_class_3>, '4': <function ClassifierWidget.create_selector_widget.<locals>.set_class_4>, 'O': <function ClassifierWidget.create_selector_widget.<locals>.toggle_selection>, 'P': <function ClassifierWidget.create_selector_widget.<locals>.toggle_selection>, 'T': <function ClassifierWidget.create_selector_widget.<locals>.run_classifier>, 'V': <function ClassifierWidget.create_selector_widget.<locals>.toggle_selection>}, key='T', func=<function ClassifierWidget.create_selector_widget.<locals>.run_classifier>, overwrite=False)
  274     key = normalize_key_combo(key)
  276 if func is not None and key in keymap and not overwrite:
  --> 277     raise ValueError(
      trans = <napari.utils.translations.TranslationBundle object at 0x7fc860917d60>
      key = 'T'
  278         trans._(
  279             'key combination {key} already used! specify \'overwrite=True\' to bypass this check',
  280             deferred=True,
  281             key=key,
  282         )
  283     )
  285 unbound = keymap.pop(key, None)
  287 if func is not None:

ValueError: key combination T already used! specify 'overwrite=True' to bypass this check

[jluethi]

Added 'overwrite=True', seems to fix the issue. Pre-existing selection & prediciton layers are overwritten with a blank one again