Specification of overhead per worker and memory allocation for dask app causes error in low resource machines

[jackransomlovell]

When using the train-pca CLI for the out-of-the-box small dataset example I ran into the following error:

$ moseq2-pca train-pca -i aggregate_results/ -o _pca_tst/ --config-file config_tst.yaml
...

File "/Users/jacklovell/opt/anaconda3/envs/moseq2-app/lib/python3.7/site-packages/moseq2_pca/util.py", line 548, in initialize_dask
    mem_limit = max(1, max_mem / nworkers)
ZeroDivisionError: float division by zero

After investigating, this is caused by allocating 0.4 of the max memory available (line 537), making the max number of workers 0 (which is set to nworkers). The error thrown was also a bit informative.