Add support for discrete demography

[molpopgen]

There are a lot of good ideas here, but they are hidden within a lot of OO complexity that is trying too hard to have 1-to-1 correspondence between C++ and Python.

To make progress:

• Get rid of PopulationConfigurationChange. Replace with a single, simple struct, MassMigration plus helper functions.
• Likewise with PopulationState and PopulationStateChange
• Likewise with the growth rate classes.
• Get rid of PopulationConfiguration class and write a function that works via the parental metadata. It should be a template that'll work with any duck-type of the current diploid metadata structure.

Also need to:

• Decide how to set up initial individual metadata? This will be done directly via the numpy interface to metadata.
• Decide how initial metadata can or cannot be used to define the max number of subpopulations.
• Decide how to parameterize migration in the new class structure.
• Implement migration in the C++ test simulation
• Add tests for migration
• Allow for migration rates to be non-negative floats
• Allow for migration rates to be "unscaled" with w/respect to N_i^t > 0.
• Update simulation code in tests to account for selfing
• Change variable in SetSelfingRate from F to S.
• Add GPL headers to all source files that survive.
• DiscreteDemography Python class should support input from a numpy array in order to provide backwards compatibility with the current API.
• Handle changes in selfing per deme over time.
• Move back-end code from test suite to fwdpy11 headers after testing.
• Fix an interface issue with MassMigration events. If there are > 1 from the same deme to different demes, then the fractions moved affected are interpreted with respect to who is left over after the preceding event. In other words, if you want to move 50% from 0 to 1 and 50% from 0 to 2, the current result is 50% to 1 and 25% to 2.
• MassMigration events need to affect growth rates/onset times/onset sizes, etc., in a manner consistent to SetDemeSizes.
• Add a constant that means "no growth"
• Remove coupling of DiscreteDemography and the ranges class by moving range types into the former. Provide a function to make sure ranges make sense vis a vis pop.generation.
• Add a placeholder documentation page that at least notes important things to fill in later.