Allow redirecting from login form directly to the /saml/login route

Uhm... I've kept going back and forth on several considerations while thinking about this patch. To keep the story somewhat short, my current state of mind is:

• This is a viable addition to the module. (I felt some resistance because 1) many people won't want this because they have some privileged users needing to log in locally; 2) the "it can be achieved by other modules" comment in the README. But really the redirect from this one login screen, belongs in this module.)
• I should not have added the 'login_link_show' boolean config option on top of the string/label; instead, changed it to just the label and just not showing it when empty. I can still drop it while maintaining backward compatibility, and will likely do so.

FWIW your option to have a default value for the label is generally viable, but I don't want to make that change (thereby making the link show up for everyone who upgrades) in a minor version update.

When I drop this checkbox, there can likely be a new boolean option, where TRUE (not FALSE) means redirect, and the #states stuff can influence the 'drupal_login_roles' setting.
(I don't know if a new boolean option is better or worse than having a special value in 'drupal_login_roles' that means "never show the login screen at all". Probably better.)

Will think about it again after finishing some other changes.

As said earlier:

• removed (bool) 'login_link_show' setting
• added (bool) 'login_auto_redirect' setting, whose use is inverted, and not dependent on / entangled with the 'login_link_title' setting. (This also means the login link is still visible on login blocks.)

I'm likely to check if I can exchange RedirectUserLoginFormSubscriber for a subscriber that sits on RoutingEvents::ALTER instead, and just execute the login code + redirect from /user/login directly to the IdP without going through /saml/login in between.

I don't know where the current test failure comes from. It's possible that it's a failure in the 8.x-3.x branch. I have to fix my somehow-broken ability to run PHPUnit locally.

Merged. I did mess up the tests a few days ago - now fixed.

My remark about "exchange RedirectUserLoginFormSubscriber for a subscriber that sits on RoutingEvents::ALTER instead" is senseless; the better solution is to have RedirectUserLoginFormSubscriber immediately do the saml/login stuff and redirect to the IDP.

I first need to fix #3211536: Reimplement response caching on login/logout routes for that, though.

I was just looking for that feature. When is it planned to be released?