benchmarks for sourmash-bio/sourmash gather implementations

See: sourmash-bio/sourmash#3232

Use environment.yml to create the basic conda environment, and then customize from there.

Make targets

make check -- run snakemake -n to see what will be done

make clean -- remove CSV files (but not rocksdb index)

make clean-all -- remove CSV files and RocksDB => rerun from scratch

Run interactively

Allocate node:

srun -p bmh --time=48:00:00 --nodes=1 --cpus-per-task=64 --mem=40GB --pty /bin/bash

snakemake -c 64

Run in batch mode

sbatch run-sbatch.slurm <conda-env>

Summarize results to Markdown

./summarize-benchmark-to-md.py  benchmarks

will produce the following Markdown output:

| prefix                  |      s |   max_rss |
|:------------------------|-------:|----------:|
| fastmultigather_rocksdb |  120.5 |       0.5 |
| fastgather              |  146.7 |      13.1 |
| fastmultigather         |  476.4 |      12.9 |
| pygather                | 2783.4 |      13.8 |

Cost of RocksDB indexing: 4:46:17 / 17177s / 14.3 GB