Openfisca-survey-manager

* For `compute_aggregate`, replace the warning by an assert when `weighted = True` (the default) and no weight is defined for the variable. Before, a warning was displayed and the aggregate was computed using uniform weights.

* Improve support for parquet file format :
- If a parquet file is provided to build-collection it will be cleaned an added to the collection as a parquet file.
- If the option `keep_original_parquet_file` is passed the parquet file is kept and added to the collection as is as is.

* New features
- Support parquet file format :
- if a parquet file is provided to build-collection it won't be converted in HDF5 but added to the collection as is.
- It is possible to provide a folder containing many files : each files will be used as a separate table.
- Run a simulation on a part of the input datasets (for the moment it works only for two entities simulations) :
- With a batch size option
- With a filter_by option

* Technical changes
- Some arguments for calibration were not consistently named.

* Technical changes
- Add config files directory option in `set_table_in_survey`

* Technical change
- With the security enforcement of PyPi, we have to use token instead of login/password.