Add a sequential reader

[rom1504]

Useful

• as a benchmark
• When data is local and doesn't benefit from parallelization
• when the bottleneck is memory rather than time

[rom1504]

this can still be helpful

[rom1504]

parallel_pieces=1 seems to do most of what we want