Allow fields to occur in multiple constructors

[andy-morris]

Currently, if multiple constructors of a datatype have a field, then multiple copies of the instances are generated.

For example:

{-# LANGUAGE TemplateHaskell #-}
{-# LANGUAGE DataKinds #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE MultiParamTypeClasses #-}
module E where

import Data.OverloadedRecords.TH
import Data.Default.Class

data Foo = A {_size :: Int} | B {_size :: Int}

pure []
overloadedRecord def ''Foo

---

E.hs:15:1: error:
    Duplicate instance declarations:
      instance Data.OverloadedRecords.ModifyField "size" Foo Foo Int Int
        -- Defined at E.hs:15:1
      instance Data.OverloadedRecords.ModifyField "size" Foo Foo Int Int
        -- Defined at E.hs:15:1

E.hs:15:1: error:
    Duplicate instance declarations:
      instance Data.OverloadedRecords.HasField "size" Foo Int
        -- Defined at E.hs:15:1
      instance Data.OverloadedRecords.HasField "size" Foo Int
        -- Defined at E.hs:15:1

This patch fixes this, by keeping track of the instances already made. While doing this, it also checks that multiple fields don't map to the same thing. In the above example, if the constructors instead had fields aSize and bSize (both mapping to HasField "size" ...), then that would be an error, even though they have the same type, since it may lead to surprising behaviour.

A possible alternative approach could be extending DeriveOverloadedRecordsParams to specify what to do in case of conflicts. But just bailing is obviously much simpler and probably(?) usually the right answer, so that's what I went for.

[trskop]

Thanks.