Add fast method for count/sum of view of BitArray

[jakobnissen]

The dispatch to this method is not ideal, because it does not handle multi- dimensional views, nor discontiguous views.
Ideally, we would have an internal API that, given a view, would iterate all contiguous sub-views. Then this new method could be called for all the conti- guous chunks.

Benchmarks

julia> using BenchmarkTools

julia> x = BitArray(rand(Bool, 1000, 1000));

julia> @btime sum(view(x, 43:849212)); # before
  251.198 μs (4 allocations: 128 bytes)

julia> @btime sum(view(x, 43:849212)); # after
  747.024 ns (4 allocations: 128 bytes)