Resolve instance variable type when assigned only in an ancestor class's method

[apiology]

This PR was written by Claude Code on behalf of @apiology.

Fixes castwide#1261

Pin::BaseVariable#visible_at?/#visible_in_closure? treats every variable as lexically scoped, walking the closure nesting of the reading site and giving up once it reaches an enclosing Namespace pin with no matching binder namespace. ApiMap#get_instance_variable_pins already walks the superclass chain to gather ivar-assignment candidates from ancestor classes, but var_at_location discarded those candidates anyway, because a descendant class's namespace is never lexically nested inside its ancestor's.

Pin::InstanceVariable now overrides visible_in_closure? to accept a pin once the lexical walk bottoms out at a Namespace with no match, since get_instance_variable_pins has already restricted candidates to the reading namespace and its ancestors. It also overrides visible_at? so the same-file requirement only applies when presence (flow-sensitive narrowing) is actually set - a plain, unnarrowed ivar assignment can legitimately live in a different file than the read (e.g. an ancestor class defined in its own file).

Example that now typechecks cleanly at --level strong:

class WidgetBase
  def run
    @name = 'set-in-run'
  end
end

class Widget < WidgetBase
  # @return [void]
  def test_no_block
    run
    puts @name.upcase
  end
end

Full spec suite: 1627 examples, 0 failures.

[apiology]

Claude: Reopened upstream as castwide#1345, from the same branch. Closing this copy.