Deriving clauses of Ptr are not calculated correctly

[endgame]

In amazonka-emr-containers, there is the following set of mutually-recursive shapes:

• Configuration contains ConfigurationList and a Sensitive field.
• ConfigurationList contains Configuration

Under #862 , the map of shapes was populated in such a way that ConfigurationList held a Ptr to Configuration, which meant that it generated a deriving clause containing a static set of class names, including Read. This fails to compile, as Configuration contains a Sensitive field, and so cannot have a Read instance.

The correct fix is to determine the set of deriving clauses by constraint solving — endgame@bdcba74 was an attempt at this, but I am abandoning it after a week and a half because it's still broken. Supporting #872 will require a serious re-architecting of the generator anyway, so let's do it properly:

• Build out a botocore library which is a direct representation of the data files provided by AWS - service, paginators, waiters, and the new endpoint-rule-set type. botocore will need some sort of "JSON deep-merge" functionality to support the splicing in of annexes. Use waargonaut and insert-ordered-containers to match the order of arguments in structs with the order they appear in botocore.
• Rewrite the generator to use botocore, and differentiate each stage (classifying shapes; applying renames/overrides; generating code)
• Consider a more modern lib than haskell-src-exts for code generation (maybe ghc-exactprint?)
• Kill the cofree comonad of shapes once and for all. it was a defensible idea when the inclusion graph of shape definitions was a DAG, but that has not been true for a very long time. If every shape reference is handled as if it were a Ptr by looking it up in a MonadReader context, edge cases like this will stop happening.

[endgame]

A possible workaround could be to allow the use of "underive" without "replace"?