Implement Identifiable Aborts (take 2)

[real-or-random]

Implement the idea from #9 (comment), ideally in a PR so we can judge the complexity.

Steps:

• Coordinator should probably check the messages received as far as possible (Should the coordinator check inputs as far as possible? #29) and raise InvalidContributionExcpetion
• The actual implementation of the idea (Identifiable aborts #52, Python investigation renames #75)
• Document exceptions properly (Identifiable aborts #52, python: Replace SessionNotFinalizedError by "Faulty" Errors #62, (Hopefully) final exceptions cleanups  #65)
• Document blame functions (chilldkg: add API docs for blaming functions #72, Python investigation renames #75)
• Change BIP text to explain IA (this needs to say that the coordinator is fully trusted for IA) (text: Explain usage aspects of blaming #71, text: Explain investigations #74)

[real-or-random]

One natural way to do this is to let each coordinator_step return additional blame data, separate for each participant. This could be an easy way of making IA optional: Just skip computing and sending this data around.

Not sure if it's worth making IA optional, but at least the additional complexity over non-optional IA is tiny.

[real-or-random]

One natural way to do this is to let each coordinator_step return additional blame data, separate for each participant. This could be an easy way of making IA optional: Just skip computing and sending this data around.

Having separate blame data also has the advantage that "main" CoordinatorMsg remains the same for everyone. That is, if you don't need IA, you can get away with the coordinator sending the exact same bytes to everyone (without wasting communication), which may be simpler.