Create zarr-only and zarr + dask benchmarks of trajectory reading

[ljwoods2]

To get a baseline of zarr's optimal performance with h5md and zarrmd files, benchmark reads and writes

1. only using zarr and
2. using zarr and dask (in something like a "lazy timeseries"

Choose a benchmark which will force dask to execute lazily-evaluated reads, like computing some kind of average. Run zarrtraj with h5md and zarr files using the same data and chunking patterns to get a fair comparison.

[ljwoods2]

Closed by #53