Optimize mapMonotonic by using an Fmap class instead of Functor

[konsumlamm]

Resolves #55.

This is the same optimization as in #59. Due to the recursive nature of our instances, the Functor instance built a lot of unnecessary <$s, which are avoided by the new Fmap class. This is the only such remaining instance that I found.

This change improves the performance of mapMonotonic significantly (see benchmarks below).

Benchmark results

Benchmarking mapMonotonic (+ 1) on a MinQueue Int with n random elements.
Before:

All
  mapMonotonic
    1000:    OK
      10.8 μs ± 286 ns
    10000:   OK
      129  μs ± 7.5 μs
    100000:  OK
      3.79 ms ± 230 μs
    1000000: OK
      98.9 ms ± 6.8 ms

After:

All
  mapMonotonic
    1000:    OK
      6.69 μs ± 398 ns, 37% less than baseline
    10000:   OK
      76.6 μs ± 2.4 μs, 40% less than baseline
    100000:  OK
      3.13 ms ± 212 μs, 17% less than baseline
    1000000: OK
      85.0 ms ± 6.0 ms, 14% less than baseline

[treeowl]

Good catch!