Import '(<>)' from Semigroup, not Monoid

[chshersh]

For example, it's not very convenient to use Semigroup.<> when you need to append NonEmpty a.

[neongreen]

We already had (<>) from Semigroup in Universum and then we changed it because it was too much hassle ;)

[chshersh]

@neongreen Can you give an example of that? Probably I didn't face this issue...

[neongreen]

IIRC there were some types without Semigroup instances and so we had to import the Monoid (<>) fairly often. I don't remember which types those were.

[int-index]

We should've added Semigroup instances upstream, since it's in base

[neongreen]

Meanwhile I have found the relevant commit and it looks like there were only six modules where we had to hide (<>) from Universum. We are also using NonEmpty more often now so I guess I'm fine with exporting (<>) from Semigroup instead of Monoid provided that we also export mappend.

[chshersh]

We can just use mappend if we need Monoid.
When Semigroup => Monoid lands in GHC 8.4.1 it will become much easier to migrate. Especially if we try by ourselves to use Semigroup instead of Monoid as much as possible.

[gromakovsky]

IIRC, most annoying example was some famous type from optparse-applicative.

[chshersh]

@gromakovsky This data type has instance Semigroup in last version of optparse-applicative.