Core: Throw a better error when documentElement is missing

[mgol]

This is a resubmit of gh-439 that was lost during the switch from JSHint + JSCS
to ESLint

Ref gh-442
Ref gh-439

(cherry picked from commit 7d92424)

[mgol]

@gibson042 I noticed that when gh-439 is applied in the jQuery repository on the 3.4-stable branch, it causes a failure of the contents() for <template /> test. This is because when find is run on an element inside of <template>, setDocument sets its document to its enclosing document fragment which has a null documentElement. Do you still think this change is fine or should we not land it in the end?

cc @dangkyokhoang

[gibson042]

We discussed in-meeting and had rough consensus that the right behavior was to stop at a disconnected fragment but to traverse up through a fragment's <template> host when one exists. But unfortunately, it seems difficult and maybe impossible to differentiate the two. My ideal change would instead be to tolerate null documentElement, but more changes will be required for that.

[mgol]

Since this PR doesn't solve the issues we have, I'm closing it, opening an issue instead: #458.