Fix logic for subclass restricted against uninstantiated nested generic superclass

[HertzDevil]

Fixes #9660.

Also affects certain flow typing constructs:

class A(T); end
class B(T) < A(T); end
class C(T) < B(T); end
class C2 < B(Int32); end

# NoReturn before this PR
typeof((x = C(Int32).new).is_a?(A) ? x : raise "") # => C(Int32)

# NoReturn before this PR
typeof((y = C2.new).is_a?(A) ? y : raise "") # => C2

# [#<C(Int32):...>] before this PR
[C(Int32).new].reject(A) # => []

# [#<C2:...>] before this PR
[C2.new].reject(A) # => []

The second commit relates to virtual generic types and fixes the following:

# false even with the first commit alone
C(Int32).new.as(A(Int32)).is_a?(B) # => true
C2.new.as(A(Int32)).is_a?(B)       # => true

# B(Int32) (non-virtual) before this PR
typeof((z = C(Int32).new || A(Int32).new).is_a?(B) ? z : raise "") # => B(Int32)+

# [#<A(Int32):...>, #<C(Int32):...>, #<C2:...>] before this PR
[A(Int32).new, B(Int32).new, C(Int32).new, C2.new].reject(B) # => [#<A(Int32):...>]

The following cases, where a superclass is restricted against a subclass, are unaffected and left to a future PR: (this part is related to #10519)

# should be B(Int32)+
typeof((x = A(Int32).new).is_a?(B) ? x : raise "") # => NoReturn

# should be C2
typeof((x = A(Int32).new).is_a?(C2) ? x : raise "") # => NoReturn

[sdogruyol]

Thank you @HertzDevil 🙏

[straight-shoota]

For reference, the last commit is about #10519 (comment)