Bypass bracketing for AbstractRanges

[DaniGlez]

Bypass the bracketstrictlymontonic stage of the index lookup for ranges in searchsorted...correlated, because the corresponding searchsorted... functions for ranges in Base already do O(1) exact lookup without searching. This results in faster interpolation when t is a range. The first linear interpolation tests are also repeated for ranges.

Benchmark:

exec_times = Float64[]
for x in 3:9 
    N = 10^x
    u = rand(N)
    t = 1:N
    li = LinearInterpolation(u, t)
    b = @benchmark $li(rand($t))
    push!(exec_times, mean(b.times))
end

Results (the jump in latency in the PR at some points is probably because of cache misses as we're interpolating random elements):

Close #187