update batch sizes to minimize idle cores

[binarymason]

Previously, batch sizes during optimization could leave quite a few idle cores. The changes in this PR ensures better cpu utilization.

Fixes #293

Before

Here are a few examples of what batch sizes would look like before this change (note number of idle cores):

length	cpu_count	minimum	before_clip	after_clip	batches	idle
4000	38	5	106	106	38	0
4000	16	5	250	250	16	0
1000	38	5	27	27	38	0
60	38	5	2	5	12	26
60	16	5	4	5	12	4

---

After

Same test dataset, but with proposed changes (minimizing idle cores):

length	cpu_count	minimum	before_clip	after_clip	batches	idle
4000	38	1	106	106	38	0
4000	16	1	250	250	16	0
1000	38	1	27	27	38	0
60	38	1	2	2	30	8
60	16	1	4	4	15	1

[kernc]

I agree, spawning extra processes, particularly by forking, is not enough an overhead to justify maintaining min. 5 jobs per core, when single runs can take long by themselves.

Thanks for finding this!