Fix after_omniauth_failure_path_for for engine-mounted Devise

[Fivell]

Summary

Fixes after_omniauth_failure_path_for for hosts that mount Devise inside a Rails engine and pin URL helpers to it via Devise.router_name = :<engine_name>.

Why it was broken

after_omniauth_failure_path_for(scope) was calling public_send(:"new_#{scope}_session_path") directly on the controller. That works when the gem's session route is mounted on Rails.application.routes (the default), because controllers include the main_app url helpers. But for engine-mounted hosts the gem mounts the route inside <Engine>.routes, so the concrete helper lives on the engine proxy (e.g. console_engine.new_admin_user_session_path), not directly on the controller — calling it raised:

NoMethodError: undefined method `new_admin_user_session_path' for an instance of ActiveAdmin::Oidc::Devise::OmniauthCallbacksController

Devise's own new_session_path(scope) doesn't help either — it's only generated when :database_authenticatable is in Devise::Mapping#used_helpers, and OIDC-only models never load that module.

Fix

Replicate Devise's URL-helper dispatch (see devise/controllers/url_helpers.rb#generate_helpers!) — look up the mapping's router_name, send it on the controller to get the engine proxy (or fall back to self for main_app), then public_send the concrete session helper on that context:

def after_omniauth_failure_path_for(scope)
  router_name = ::Devise.mappings[scope].router_name
  context     = router_name ? send(router_name) : self
  context.public_send(:"new_#{scope}_session_path")
end

Test plan

• rake spec:all is green — covers default, engine-mounted, and isolated-engine setups.
• Verify on a host with Devise.router_name = :<engine> that an OmniAuth failure redirects to the SSO landing page instead of 500-ing.