Schedulers as structs

[carstenbauer]

This PR introduces structs for different schedulers. This enables us to have different options (fields) and default values for different schedulers (#22).

Notes:

• The static and dynamic schedulers have nchunks but the greedy scheduler has ntasks, because it doesn't do any chunking.
• We default to nthreads() tasks for static and greedy but 2*nthreads() tasks for dynamic to have a bit of load balancing by default. (unclear if this is the best choice, see questions below)

Questions:

• Should the default for DynamicScheduler be nchunks = 2*nthreads() or nchunks = nthreads(), i.e. load balancing or not?
• Maybe we should choose the latter and have a separate LoadBalancingScheduler as an alias for DynamicScheduler(nchunks=2*nthreads)?
• Should we export all scheduler structs? (I currently do)

Closes #22

[carstenbauer]

Documentation preview: https://juliafolds2.github.io/OhMyThreads.jl/previews/PR45/

[MasonProtter]

Yep, it all looks good to me!

Regarding the questions:

1. I typically like 2*nthreads() as the default for dynamic. I think it makes the most sense in almost all cases, and then people can tune from there if they have a problem where those extra tasks really are expensive. IMO 1*nthreads() only really makes sense if your workload is highly uniform, or you're right on the cusp of when multithreading is not profitable (order 10-100 microseconds or whatever)

2. I think the alias there would be kinda misleading. The greedy scheduler is much more of a load balancer than just oversubscribing.

3. If people don't want them in their namespace, they can just choose what to import.