Use verify_authenticity_token directly

[nevans]

This is a significant refactoring, in order to use verify_authenticity_token directly.

This was motivated by issues with GitHub CodeSpaces: their proxy mangles the Origin header (infuriating). That can be hard to debug already, but unfortunately this gem lost all of the context for why verified_request? returned false. Fortunately, verify_authenticity_token does add the needed context to the exception message. Once I could see that message, my issue became obvious.

I believe this should be mostly safe, because the verify_authenticity_token method name is the primary public API through which users interact with ActionController::RequestForgeryProtection (e.g: skip_before_action :verify_authenticity_token). There is some risk that rails may change its implementation such that we need to update to this class to continue supporting it. However, I believe that risk is small: this code should work for rails 4.2 through 8.0, even though ActionController::RequestForgeryProtection has been through significant changes and refactorings during that time.

In order to use verify_authenticity_token, we need a working #logger and we need to ensure that verify_authenticity_token raises an exception.

The #logger is simply delegated to OmniAuth.logger.

I converted TokenVerifier.config to work more similarly to the standard rails approach for inheriting a configuration object (should this be split into its own PR?). ActionController::RequestForgeryProtection adds its defaults to TokenVerifier's config when it is included. By deleting all of the keys on our local config object, we ensure that every config setting is inherited from the parent (which is is ActionController::Base.config). By using the inheritable config (rather than simply overriding the methods) we gain the ability to override the ActionController::Base config for TokenVerifier. I did this for two config values:

• If forgery_protection_strategy isn't configured to raise exceptions, then verify_authenticity_token doesn't raise an exception, and OmniAuth can't detect any failure when it's called.
• Both OmniAuth and ActionController::RequestForgeryProtection will log the failure, so it's should be safe to set log_warning_on_csrf_failure to false.

Additionally, I wrapped the exception in OmniAuth::AuthenticityError, so it would be caught by the appropriate OmniAuth rescue clause:

    rescue OmniAuth::AuthenticityError => e
      fail!(:authenticity_error, e)
    end

[nevans]

FYI: I rebased this on top of #22, so it would pass CI against all versions of rails.