Gitlab/StrongMemoize: Remove forward declaration of `strong_attribute_attr`

Problem

In https://docs.gitlab.com/ee/development/utilities.html#strongmemoize we suggest two form of declaring strong_memoize_attr:

• 1️⃣ "post" declaration
• 2️⃣ "forward" declaration

class Find
  include Gitlab::Utils::StrongMemoize

  # :one:
  def result
    search
  end
  strong_memoize_attr :result

  # :two:
  strong_memoize_attr :enabled?, :enabled
  def enabled?
    Feature.enabled?(:some_feature)
  end
end

1️⃣ is commonly used in Ruby for other access modifiers like private and protected:

class Foo
  def foo
  end
  private :foo
end

However, 2️⃣ won't work for access modifiers:

private :foo
def foo
end

__END__
Traceback (most recent call last):
        1: from foo.rb:1:in `<main>'
foo.rb:1:in `private': undefined method `foo' for class `Object' (NameError)

For this reason it's surprising to see that strong_memoize_attr supports this style.

Proposed solution

Remove the ability to declare strong_memoize_attr before the method definition.

This following the code should raise NameError similar to what access modifiers are doing:

  # :two:
  strong_memoize_attr :enabled?, :enabled
  def enabled?
    Feature.enabled?(:some_feature)
  end

"Inline" declaration

In theory the following code will also work:

strong_memoize_attr def foo
end

But it seems this is not something we prefer in the codebase.

Discussion

The following discussion from !102915 (merged) should be addressed:

• @splattael started a discussion: (+1 comment)

  Related to !102915 (comment 1165797188)

  Opened a new thread to have more room to discuss.

  I only now saw that we allow defining strong_memoize_attr before the actual method definition. That's neat but also a bit sneaky 🤓

  https://gitlab.com/gitlab-org/gitlab/-/blob/master/spec/lib/gitlab/utils/strong_memoize_spec.rb#L38

  Although it works I am wondering if queuing is actually needed because it's surprising and does not match the behavior of other modifiers like private and protected.

  private :foo
  def foo
  end
  
  __END__
  Traceback (most recent call last):
          1: from foo.rb:1:in `<main>'
  foo.rb:1:in `private': undefined method `foo' for class `Object' (NameError)

  Suggestion (non-blocking) Thoughts on removing this feature from Gitlab/StrongMemoize?

  I'll create a follow-up.