Study what `SLiM` does with metadata vs. tskit C/Python API

[gregorgorjanc]

See

tskitr/R/inst/STATE_and_AIMS.md

Line 170 in 88985e1

One downside of using `reticulate` is the overhead of calling Python functions, which can add up if they are called many times, say within an R loop. This overhead is minimal for most analyses as these would call a few functions, so most of the work, looping, etc. is done on Python side, often in `tskit`'s C code). However, the overhead can be limiting for repeated calling of `tskit` functions, say for tree sequence recording and bespoke analyses (though ideally these would be implemented in Python and possibly ported to C/C++). Importantly, metadata functionality is only available in Python, though `SLiM` does something of it's own (TODO: study what `SLiM` does with metadata).