Performance of struct ACSets

Writing down this issue so that we remember that there were a couple of small performance regressions with struct acsets, and it would be good to see if we can fix them later on.

Here are the benchmarks for struct acsets vs. the old implementation; they are a useful starting point for future investigation.

https://github.com/AlgebraicJulia/Catlab.jl/pull/376/checks?check_run_id=3325008558