implement a preliminary version of pairwise reduce

[exaexa]

this needs some improvement but looks to be the way to reduce a lot of Value's.

[exaexa]

cc @stelmo opinions welcome

[exaexa]

stupid minibenchmark even with the completely type-broken version:

julia> x = C.LinearValue.(([i] for i=1:1000000),Ref([1]));

julia> @time sum(x)
  9.970770 seconds (10.11 M allocations: 20.243 GiB, 8.99% gc time)
ConstraintTrees.LinearValue([1, 2, 3, 4, 5, 6, 7, 8, 9, 10  …  999991, 999992, 999993, 999994, 999995, 999996, 999997, 999998, 999999, 1000000], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0  …  1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0])

julia> @time C.preduce(+, x, init=0)
  1.568678 seconds (26.34 M allocations: 1.670 GiB, 6.86% gc time)
ConstraintTrees.LinearValue([1, 2, 3, 4, 5, 6, 7, 8, 9, 10  …  999991, 999992, 999993, 999994, 999995, 999996, 999997, 999998, 999999, 1000000], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0  …  1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0])

[exaexa]

ok good I de-haskellized this and

julia> @time C.preduce(+,x)
  0.632748 seconds (4.34 M allocations: 916.133 MiB, 6.17% gc time)

I'll merge this into the other perf-related MR to simplify testing