Birdhouse-finch

===================
* Fixed iter_local when depth > 0 to avoid all files to be considered twice
* Revised documentation configuration on ReadTheDocs to leverage Anaconda (Mambaforge)
* Minor adjustments to dependency configurations
* Removed configuration elements handling from `finch start`. One can still pass custom config files, but all configuration defaults are handled by `finch/default.cfg` and the WSGI function. `jinja2` is not a dependency anymore.

===================
* Generalize ensemble datasets configuration.
- Datasets usable by ensemble processes are now specified through a YAML file which is pointed to in the configuration.
- As a consequence processes are generated according to the available variables. Similarly for the allowed values of some inputs on these processes.
- The output name now includes the dataset name (if a custom name was not specified).
- ``finch.processes.xclim`` was removed, there is no static module of processes.
- Input "rcp" has been renamed to "scenario".
- Input "dataset_name" has been fixed and renamed to "dataset".
* Update to xclim 0.38.0.
* Improved subset_grid_point_dataset & subset_bbox_dataset performance when using local files.

==================
* Fix Finch unable to startup in the Docker image.

==================
* Avoid using a broken version of ``libarchive`` in the Docker image.

==================
* Fix use of ``output_name``, add ``output_format`` to xclim indicators.
* Change all outputs to use ``output`` as the main output field name (instead of ``output_netcdf``).
* Updated to xclim 0.37:

- Percentile inputs of xclim indicators have been renamed with generic names, excluding an explicit mention to the target percentile.
- In ensemble processes, these percentiles can now be chosen through ``perc_[var]`` inputs. The default values are inherited from earlier versions of xclim.
* Average shape process downgraded to be single-threaded, as ESMF seems to have issues with multithreading.
* Removed deprecated processes ``subset_ensemble_bbox_BCCAQv2``, ``subset_ensemble_BCCAQv2`` and ``BCCAQv2_heat_wave_frequency_gridpoint``.
* Added ``csv_precision`` to all processes allowing CSV output. When given, it controls the number of decimal places in the output.

==================
* Preserve RCP dimension in ensemble processes, even when only RCP is selected.
* Pin ``dask`` and ``distributed`` at ``2022.1.0``, see https://github.com/Ouranosinc/PAVICS-e2e-workflow-tests/issues/100