Ensure validation is respected by containing form

[rnicholus]

Over the past few months, I've spent quite a bit of time on ajax-form. At this point, it seems to work pretty well. It "submits" all native form fields (via XHR), respects enctype, action, and method attributes common to a bare <form>, and adds a bunch of other stuff that native HTML forms don't offer. Recently, I added support for submission of custom element fields, such as core-input (and related fields), file-input, paper-dropdown-menu, and any other custom element form field with a name attribute and a value property. While validation has been respected on native form fields for some time, I've struggled to respect validation on custom fields. As far as I can tell this is actually not an issue specific to ajax-form. Rather, it affects all form fields inside of any form, provided the form fields are CE with shadowed inputs.

Unfortunately, the underlying form apparently cannot see that a form field is invalid if it exists in the shadow DOM. This is really tricky to support, but I seem to have come up with a method. The solution here, to ensure that invalid shadow DOM form fields prevents form submission, is to create a "hidden" delegate input in the light DOM that mirrors the validity of the shadowed form field. Another benefit of this approach is that the browser (if supported) will also render a pop-up with the customValidationMessage when the form submit is held up due to an invalid CE form field. If this "hidden" delegate is positioned directly next to the CE, the pop-up seems to target the CE itself.

I have made appropriate changes in another Polymer CE I created, file-input in rnicholus/file-input#19, and that seems to work well with ajax-form now, in terms of validation support. Locally, I've also made the proper adjustments to core-input to ensure validation is respected by a <form> during the submit attempt.

Respecting validation of CE form fields is currently a big issue (IMHO) and not yet solved by any other method (that I am aware of). My solution, while a bit unsettling, appears to be the only way to tackle this, as far as I can tell.

I'm wondering what your thoughts are on this. I'm certainly open to discuss further, and am even willing to open up a pull-request with my changes to core-input (which likely will elicit further discussion).