CG: semantics of automatic inference of constraints / 'implements' declarations

[vasslitvinov]

main issue: #8629     whether to infer #16952

If/when we want to infer a constraint / an implements declaration at a callsite of a CG function according to #16952, what are the rules? In particular:

(equiv) A constraint is satisfied implicitly IFF if we place the corresponding implements statement with an empty body immediately before the callsite, it would be correct.

interface I {
  proc f(arg: Self);
}

proc cgFun(arg: ?Q) where Q implements I {
  f(arg);
}

// The below call cgFun(5) resolves IFF
// the following implements-statement would be correct:
//int implements I;

cgFun(5);

(no-shadow) If there is a matching implements statement that is visible at the callsite, no inference takes place. That is, we do not use any shadowing functions even when they are available. For example:

interface I {
  proc f(arg: Self);
}

proc f(arg: int) { writeln("f#1"); }
int implements I;  // uses f#1

proc cgFun(arg: ?Q) where Q implements I {
  f(arg);
}

{
  // f#2 shadows f#1
  proc f(arg: int) { writeln("f#2"); }

  // If we inferred `int implements I` here,
  // it would be implemented with f#2.
  // However, since there is a visible implements-statement already,
  // we use the implementation that comes with that, which is f#1.
  cgFun(5);
}

[lydia-duncan]

A constraint is satisfied implicitly IFF if we place the corresponding implements statement with an empty body immediately before the callsite, it would be correct.

I might rephrase this as "A constraint is satisfied implicitly IFF a corresponding implements statement with an empty body could correctly be placed immediately before the callsite"