Improve speed of shannon and ginisimpson

[barucden]

Hi. I noticed that some computations could be saved in the definition of shannon and ginisimpson.

I am not a user of this package, so I am not sure about the impact. However, some measurements:

julia> const x = rand(1000);

julia> @btime shannon(x)
  148.720 μs (2 allocations: 15.88 KiB)
6.730219402261953

julia> @btime new_shannon(x)
  11.158 μs (0 allocations: 0 bytes)
6.730219402261952

julia> @btime ginisimpson(x)
  4.278 μs (2 allocations: 15.88 KiB)
0.9986907680545117

julia> @btime new_ginisimpson(x)
  296.874 ns (0 allocations: 0 bytes)
0.9986907680545117

The speed difference should increase with the length of the input.

Be aware that the result of the new shannon is little bit off due to numerical accuracy.

[kescobo]

🤔 Ack! Sorry for this languishing. I could have sworn I replied to this and merged it already... 😱

Anyway, looks great - these functions are rarely bottlenecks, but more speed with no cost is obviously great. Thanks for the contribution!