[m-rev.] for review: improve instance diagostics

[Julien Fischer]

On Wed, 27 Jun 2018, Zoltan Somogyi wrote:

> For review by anyone. I hope this will help people to understand
> the messages for the problem reported yesterday on m-users.

> Improve diagnostics for bad instance declarations.
> 
> compiler/check_typeclass.m:
>     When a type is an instance declaration is not a type constructor,
>     don't just say what it is; say also what it should be.

s/is an/in an/

>     When a type is an instance declaration is a type constructor but some

Ditto

>     of its arguments are not type variables, don't just say so: say *which*
>     of their arguments are the problem.
>
>     Do not use "argument" to refer to both "the arguments of the instance
>     declaration" and the "arguments of a type constructor"; use the terminology
>     "instance type" for the former.
>
>     Always end the line after the ":" in the error's top level description.
>
>     Omit the module qualification of the typeclass name in error messages,
>     since the base name of the typeclass, and the context, should always
>     be enough to eliminate any ambiguity.
>
>     Fix a bug in the use of choose_number.
>
>     Don't gather the set of seen types, since it was never used.

That looks fine otherwise.

Julien.