While investigating ways to simplify and improve placeChild I stumbled upon a bug in how we diff Fragments. If child of a Fragment is unmounted, a Fragment's _nextDom pointer could point to this unmounted child. In situations where this happens, the Fragment's parent will correct this by re-appending all the children after the Fragment. While the DOM output is correct, the way we go about achieving it is sub-optimal. This change fixes the _nextDom to not point at a DOM that is going to be unmounted.

For example, take the following code:

<div>
  <Fragment> <!-- Fragment 1 -->
    <div>1</div>
    <div>2</div> <!-- this node will get unmounted -->
  </Fragment>
  <Fragment> <!-- Fragment 2 -->
    <div>A</div>
  </Fragment>
</div>

If doing a diff from the top where <div>2 gets unmounted, Fragment 1 should finish it's diffChildren call with a _nextDom pointer to <div>A. However, before this fix, when Fragment 1 diffs <div>1, it sets its _nextDom pointer to <div>1.nextSibling which is <div>2 (it hasn't been unmounted yet). This fix detects that the _nextDom pointer will be unmounted and resets it to the next DOM sibling, in this case <div>A.

This PR fixes a long standing Fragment issue that prevented us from asserting expected DOM operations on Fragment diffing and this PR enables us to them turn on! We can now catch regressions in how we mount & move around fragment children!

Specifically, on master the test "should handle changing node type within a Component that returns a Fragment #1326" in fragments.test.js has more than necessary dom operations. The same is true for the tests I've added. All have reduced number of DOM ops.