Add filterA, a faster version of filterM

[hdgarrood]

Change filterM to just be filterA in order to have this be a non-breaking change; I was just thinking it would be nice for all these performance improvements to be available for people without having to update to the next major version.

This removes another use of uncons, refs #71.

The new devDependency on Const is because that seemed like the best choice for testing that effects are applied in the right order; the tests were not previously testing that filterM applies effects in the right order.

Benchmarking code:

benchFilterM :: Benchmark
benchFilterM = mkBenchmark
  { slug: "filterM"
  , title: "filterM"
  , sizes: A.range 1 5 <#> (_ * 500)
  , sizeInterpretation: "Length of the array"
  , inputsPerSize: 1
  , gen: \n -> vectorOf n (arbitrary :: Gen Int)
  , functions: [ benchFn "current"   (currentFilterM pred)
               , benchFn "suggested" (suggestedFilterM pred)
               ]
  }
  where
  pred :: Int -> WriterT Unit Identity Boolean
  pred x = tell unit $> even x

I had to stop at 2500 elements since the current implementation stack overflows at around 3k, but the new implementation should work for much larger arrays, because of the stack-safety of the Traversable Array instance. I tested this implementation with up to 10k elements and it seemed to work fine.

[hdgarrood]

Oh, nearly forgot - benchmark results!

[hdgarrood]

Also I should probably check - are we ok with deprecating filterM now that we are exporting filterA?

[paf31]

👍 LGTM, thanks (and also +1 to deprecating filterM)

[hdgarrood]

Great. :)