Unify handling of fixations.

[molpopgen]

Placeholder issue for now...

The current handling of fixations is not ideal:

• it leads to modeling errors
• the pruning policy stuff affects efficiency of fitness calculations.

Any fixes will break API...

To address this:

• [ ] Fixations should always be removed.
• [ ] Neutral/selected fixations should be separated in population objects
• [ ] Genetic value functions need updating to calculate the contribution of fixations each generation and apply that to each diploid.
• [ ] Meta-populations and multi-locus simulations may need special thought. For the latter, selected fixations will need sorting by position to match them up to each locus.

[molpopgen]

Tree sequence recording makes this a bit less pressing? See #167, for example.