Add improved searchsortedfirstcorrelated and bracketstrictlymontonic

[ChrisRackauckas]

For benchmarks see:

• Add a fast path for evenly-spaced abscissae DataInterpolations.jl#198
• Use correlated searching for repeated interpolations DataInterpolations.jl#147

While the cost of constructing the interpolator does not change, tracking the last index results in a best-case speedup of ~2.7x for CubicSpline, when successive values are close together (and a little higher for simpler interpolators). In the worst case (where successive values are always on opposite ends of the vectors), it can result in a ~15% slowdown due to the unhelpful expanding binary search at the beginning. However, the original approach of not tracking the index at all is also still available; it now involves essentially one extra if statement, which seems to be lost in the timing noise.

[ChrisRackauckas]

Note this is using some downstream tests to supplement since DataInterpolations covers the sorted search pretty comprehensively.