Remove MonadFilter, and fix MonadCombineLaws

[eed3si9n]

Fixes #1336

The MonadCombine typeclass unifies Monad, FunctorFilter, and
MonoidK, bringing three of the lineages together. This uses the
emptiness notion introduced in MonoidK and flatMap to implement
mapFilter, along with unite.

Given that MonadCombine exists (our equivalent of MonadPlus) there's no need for MonadFilter to exist.

The laws are migrated over from MonadFilter to MonadCombine, except these laws would now check identity laws and the consistencies against the correct F.empty from MonoidK.

[mpilquist]

Given that MonadCombine exists (our equivalent of MonadPlus) there's no need for MonadFilter to exist.

Can you expand on this? Doesn't this mean we can't define MonadFilter instances for monads that don't have lawful / sensible combine implementations?

[codecov-io]

Current coverage is 91.54% (diff: 88.88%)

Merging #1342 into master will increase coverage by 0.15%

@@             master      #1342   diff @@
==========================================
  Files           237        233     -4   
  Lines          3567       3550    -17   
  Methods        3502       3487    -15   
  Messages          0          0          
  Branches         64         62     -2   
==========================================
- Hits           3260       3250    -10   
+ Misses          307        300     -7   
  Partials          0          0          

Powered by Codecov. Last update 449e0b0...5e0165b

[eed3si9n]

Doesn't this mean we can't define MonadFilter instances for monads that don't have lawful / sensible combine implementations?

Yes. If I want to define filter there's now FunctorFilter, so what's the use of introducing a notion of emptiness that's different from combine?

[edmundnoble]

MonadFilter was removed in cats-mtl as @eed3si9n says because there are no laws, as well as MonadCombine because the laws are all derivable from parametricity and Alternative's laws.