This PR modifies set_params to also execute a full reset, which calls any __init__ code such as dynamic tag setting.

Also adds parameter inspection related class methods to BaseObject:

• get_param_names - retrieves list of parameter names
• get_param_defaults - retrieves name/default dict of parameter defaults

Why set_params should execute a reset: @aiwalter pointed me to this specification document in sklearn which makes a key interface assumption explicit, namely that the result of set_params must be the same as __init__(**self.get_params()).

Since sktime may have some __init__ content besides self writes (in deviation from sklearn due to varia discussed in #2573), we can ensure consistency here by calling reset from set_params.
This relates to discussion in, and also partially addresses #2573 .

As a secondary effect of this, the test_set_params test was failing in instances where parameter related logic was present in __init__, e.g., cloning estimators to component_name_, due to testing logic in sklearn's check_set_params which was called from test_set_params. The reason for that is that sklearn's test would overwrite parameters with np.inf etc to mimic a change, which is fine in sklearn since it strictly has no logic in __init__. However, sktime estimators may have logic, such as dynamic tag setting, which therefore would break the set_params now identical with __init__.

Therefore, the test_set_params test was changed from overwriting with np.inf etc, to overwriting with values from other test cases as available through get_test_params and get_param_defaults (needed for full set), which ensures validity.