Refactor: Make process of collection independent of the choice of main dictionary

[fbanados]

Currently, the process of generating an importjson file strongly depends on the contents of the CW dictionary. We intend to make the following changes:

• There's repeated code to process each dictionary. Isolate code that processes complete dictionaries in one place
• Encapsulate source specific changes into each source's class
• Separate special initialization processes currently done while converting CW into a general class so that any source can be "the first to appear"
• Turn aggregation from a global operation to a one-to-one process so that source priority can be easily changed.
• Ensure entries that have no mapping on previously aggregated entries are still included (depends on previous issues)\
• Refactor instructions to use _altlab versions of alternative dictionaries instead of main (mostly immutable) sources.
• Change documentation and dependencies on use of FSTs: Currently we only use the relaxed analyzer as a way to account for spelling differences between dictionaries.

It is currently expected that finalizing this process will expand the crkeng_dictionary.importjson file with around 8k senses that are currently being discarded by the matching process.

[aarppe]

The fields in CW and MD relevant to dictionary comparison from the LEXC perspective are discussed in #125.