Add Threaded parallel scheme

[odow]

Closes #599

This is heading in the right direction:

julia> @time simulations = SDDP.simulate(
           model,
           10_000;
           parallel_scheme = SDDP.Threaded(),
           custom_recorders = Dict{Symbol, Function}(
               :thread_id => sp -> Threads.threadid(),
           ),
       );
 14.560144 seconds (58.22 M allocations: 2.101 GiB, 5.02% gc time, 0.06% compilation time)

julia> @time simulations = SDDP.simulate(
           model,
           10_000;
           parallel_scheme = SDDP.Serial(),
           custom_recorders = Dict{Symbol, Function}(
               :thread_id => sp -> Threads.threadid(),
           ),
       );
 47.935609 seconds (58.31 M allocations: 2.106 GiB, 2.24% gc time, 0.10% compilation time)

Avoiding the various race conditions is a bit fiddly though.

TODOs

• integer stuff?

[codecov]

Codecov Report

Attention: Patch coverage is 85.91549% with 10 lines in your changes missing coverage. Please review.

Project coverage is 93.40%. Comparing base (430b56a) to head (f02dc91).

Files	Patch %	Lines
src/algorithm.jl	76.66%	7 Missing ⚠️
src/plugins/parallel_schemes.jl	88.88%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #758      +/-   ##
==========================================
- Coverage   93.44%   93.40%   -0.04%     
==========================================
  Files          27       27              
  Lines        3416     3458      +42     
==========================================
+ Hits         3192     3230      +38     
- Misses        224      228       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[odow]

So the issue is that prepare_backward_pass assumes that we will be doing only forward passes or only backward passes. We need to change this to apply per-node.

[odow]

There's still testing to be improved, but I think this is ready for basic usage.