Further exploit sparsity in computing derivatives

[chrisdembia]

See:

• Betts and Huffman 1999 Exploiting sparsity in the direct transcription method for optimal control and
• Patterson and Rao 2012 Exploiting sparsity in direct collocation pseudospectral methods for solving optimal control problems.

[chrisdembia]

This likely requires moving the differentiation code into its own class(es), so that we can compute finite differences for any function.

[chrisdembia]

A simple thing to do is for the Trapezoidal class, do not assume that initial and final time are coupled to every other variable (this is not the case in the implicit double pendulum).