Parsyfiles

* improved support for collections by supporting Mappings, Sequences, and AbstractSets instead of Dict, List, and Set
* added automatic conversion of string to numbers and booleans in jprops properties reader
* replaced dict_to_single_row_or_col_df with dict_to_df, able to handle both single-level and multi-level dictionaries
* fixed filesystem mappings : now supporting multifile objects without children (=empty folder or file without extension)
* fixed code for detecting a typing List or Set (was relying on typing.Collection that is not always available)

* Two new main functions `parse_item` and `parse_collection` at module level
* Custom parsers for MULTIFILE may now be registered (since a multifile parser is now a parser like all others)
* Added the concept of converters and conversion chains. A converter registry has an "intelligent" logic to infer what type of converters to chain in order to create value (preventing useless conversion chains, cycles, etc.)
* Added support for primitive types, yaml and numpy primitives
* Parsers and converters are defined according to the new object model for collections, objects, dataframes, configparser. (addded/fixed support for json, jprops, pickle, ...). There are now two ways to parse config files: as dictionary of dictionaries, or as a dictionary
* Added the ability to parse collections in 'lazy' mode
* Added support for Tuple and Set for collections
* added automatic conversion of item values for all dict, list, tuple and set parsers/converters
* added a system to pass options to the parsers/converters
* Improved Logging and Error handling to ease debug
* added support for classes written with `attrs` package
* Major code refactoring :
* split into several files to separate the concerns : file system / parsing chains / framework / typing and PEP484
* object model for persisted objects, converters, parsers, and parsers/converters registries
* All parsers and converters shipping with the framework are now plugins just as the custom ones : no hardcoded difference. This made the plugin mechanism quite "strange case-proof".
* Added a test file specifically for parser registry

This first version was mostly function-based and did not provide a way to parse simple objects from dictionaries.