Invest-ucm-calibration

- fix readthedocs build issues by adding more libraries to `autodoc_mock_imports`
- added docstrings for public API methods
- added `UCMCalibrator.calibrate` method (uses `Annealer.anneal`)
- fix passing `ucm_args` to `get_comparison_df` and `get_model_perf_df` methods

- custom `dates` argument in `UCMWrapper`
- shortcut to useful `UCMWrapper` methods in `UCMCalibrator`
- using `_ucm_params_dict` property to get params from annealer state
- added sample comparison and model performance methods
- added `predict_t_da` method (works with xarray)
- renamed variables `model_args` -> `ucm_args`, `DEFAULT_MODEL_PARAMS` -> `DEFAULT_UCM_PARAMS`
- default model parameters from `settings` module in `base_args` attribute of the `UCMWrapper` class

- update `base_args` with `model_args` in `predict_t_arr`
- compute R^2 with `scipy.stats` (instead of `sklearn.metrics`)
- exclude zero kernel distance to avoid nan/infinity errors

- set `natcap.invest` version requirement as `>=3.8.0`, restrict pygeoprocessing to `<2.0`
- alignment of the temperature rasters if needed
- using `src.dataset_mask()` method instead of `arr != src.nodata`
- dumped `aoi_vector_filepath` argument, automatically generating a dummy one instead (since it is not used in the calibration)

- fix automatic defaults for `num_steps` and `num_update_logs`
- set num_steps/num_update_logs to 2 in tests to test `move` method

- initial release