Specify known sparsity pattern in mixed mode Jacobian

[ErikQQY]

In the mixed-mode Jacobian example in #684, how can we specify the known sparsity pattern in the backend construction? And the related documentation on mixed mode Jacobian is really scarce, maybe we can put the benchmarking example as documentation to illustrate how to exploit MixedMode Jacobian to achieve faster Jacobian computation.

(Open a new issue to not disrupt #684)

[adrhill]

There is KnownJacobianSparsityDetector, which you can use as a sparsity_detector within a AutoSparse backend.

[gdalle]

Note however that this still won't allow you to manually specify which coefficients of the Jacobian should be computed with forward AD and which coefficients should be computed with reverse AD. The whole point of bidirectional coloring is to make this decision for you