py math: Support matrix inverse for dtype=AutoDiffXd and Expression

[EricCousineau-TRI]

Pending some form of resolution for #8116 for proper ufuncs, we should add a shim for matrix inverse in pydrake.math.

Slack convo: https://drakedevelopers.slack.com/archives/C3YB3EV5W/p1554344618003700

good question from an underactuated student:

Thanks a lot I did the same as u did in quadrotor2d! I actually have a new problem for DirectCollocation and AutoDiffXd.

I'v seen there's the open feed on how to make numpy linear algebra compatible with AutoDiffXd type arrays.

I guess np.linalg.inv is not yet compatible (I get that TypeError: No loop matching the specified signature and casting was found for ufunc inv error).

My question is: when you have a system in manipulator equations form, and you want to fill up the _DoCalcTimeDerivatives method, do you have to  invert the inertia matrix  by hand and plug in the algebraic coefficient of the inverse to make it compatible with  AutoDiffXd?

Or is there a built-in way to do this?

\cc @RussTedrake