[x] Add creation & consumption of cache to CI. See #43
[x] Increase queue size (and make it configurable/adaptive). Currently, running boshernitzan conjecture search, runs a bit low on the queue starting at about a+b+c=140. For similar a, b, c, creating a surface takes about 12 times the time than it does to check the conjecture. So we do not run at full load when running on more than 13 cores. (Alternatively, we could fill the queue in parallel, but this probably requires #42 to do properly.)
fixes #39.
Dependencies
TODO