Add benchmark for reading spike_times from units table

[oruebel]

This issue hdmf-dev/hdmf-zarr#141 discusses differences in performance when reading from an indexed column in a DynamicTable, i.e., when we have a ragged array defined via a VectorData and VectorIndex

1. spike_times = nwbfile_read.units.spike_times[:] which reads reads all values from the VectorData but does not create a ragged array
2. spike_times = nwbfile_read.units['spike_times'][:] which reads data using VectorIndex (which in turn reads from VectorData to create a ragged array

It would be useful to benchmark the difference in performance and to see if the read of ragged arrays could be optimized. This difference in performance is likely exasperated when doing remote read, because VectorIndex likely makes many read requests to VectorData to read the different segments, rather than loading the data first into memory and then segmenting the array.