Allow abstract methods to be implemented via method_missing

[kshtzkr]

Closes #11

Problem

AbstractInstanceMethodReceiver#method_missing raised AbstractMethodNotImplementedError before calling super, so a method_missing defined further up the ancestor chain (e.g. on a superclass) never got the chance to provide a dynamic implementation of an abstract method:

module MyInterface
  interface!

  abstract def demo; end
end

class Parent
  def method_missing(m, ...)
    return super unless m == :demo

    "dynamic #demo"
  end
end

class Child < Parent
  include MyInterface
end

Child.new.demo # raised AbstractMethodNotImplementedError, now returns "dynamic #demo"

Approach

As suggested in #11 (comment): call super first, and only translate the resulting NoMethodError into an AbstractMethodNotImplementedError if nothing handled the call.

Two refinements over the sketch in the issue:

• The NoMethodError is only translated when e.name == method_name. A NoMethodError for a different method — e.g. raised from inside a broken dynamic implementation of the abstract method — is a real error and propagates unchanged, instead of being misreported as "abstract method never implemented".
• The translated error is raised with cause: nil (as in the issue's sketch), since the underlying "undefined method" error is unactionable noise beneath the more precise error.

I went with rescuing NoMethodError rather than checking respond_to_missing? first, per the reasoning in the issue: lots of code overrides method_missing without overriding respond_to_missing?.

Tests

Three new specs in interface_spec.rb:

• an abstract method implemented via an inherited method_missing can be called
• an unimplemented abstract method still raises AbstractMethodNotImplementedError, with no NoMethodError cause attached
• a NoMethodError raised inside a dynamic implementation propagates as-is

All local checks pass: rake typecheck (incl. Prism), rake test (92 runs, 0 failures), rake rubocop, bin/tapioca gem --verify, bin/tapioca check-shims.

Note: #11 had an assignment request from April that was never assigned and has no linked PR — happy to step aside if that work is still in flight.