Fix initial routing state after match

[timdorr]

Continuation of #2965

#2883 was actually busted and had a test that didn't cover what it was supposed to.

In an ideal world we'd just use the same history and transitionManager for match() and on the client side, but the deprecation wrapping API messes with that too much, so this might be the best we can do for now.

[taion]

I think I have a different approach that makes it a bit more explicit. I'm really scared of going through any of these new branches when not coming from match, because that can only ever cause bugs.

[timdorr]

It's the same thing as your changes, just cleaned up and with an invariant check that explicitly tells you what's going on. I couldn't contribute to the last PR because you did it on a private fork, so this is my way of participating.

[taion]

I know, sorry – old habit from before we fixed the Travis build not to spammily run on branches. The edge case I'm still concerned with on even 5a8b362 is the case where somehow a user passes in a transitionManager but not a history, or a transitionManager plus a routing history or something like that, so I changed it to just go from taion@6ab8316 and move the router objects into a different opaque prop, but I didn't want to clobber your commits.