This change must have slipped in previously when I first introduced
the ALMaSS submodule.

(Not yet tested.)

As well as the concomitant macros, and @killother

This used to be done by Agents.jl, is now done natively

I now use parametric methods to store initialised parameter structs,
rather than creating a special registry for these with a closure.

This is a huge commit that changes much of the underlying model
architecture (which is why it all had to change at once). Agents.jl
is now no longer a dependency and the species macros have been largely
rewritten. Not everything has been adapted yet, but the model is in a
"minimally runnable" state.

Still need to write tests to make sure it works properly, but at least
it runs without errors now and seems to do what it's supposed to.

Also moved the DSL macros into a separate file. Still need to write
tests for all of this.

This commit includes a `sed` pass over all files to change all
occurrences of the former spelling "Persephone" to "Persefone":
`sed -ie "s/Persephone/Persefone/" *`

This closes issue #50

This allows us to get rid of the annoying `animal.traits["traitname"]`
syntax and replace it with the much simpler `animal.traitname`.

Using the GLOBAL_RNG introduces global state, which must be avoided
to preserve reproducibility. Therefore, all Persephone code must use
`model.rng` whenever calling `rand()`/`shuffle!()`/etc.

This seems to work as intended now! :D