[WIP] Add prescribed dilation to visco plastic

[naliboff]

Supersedes #3119, but is only open now for discussion and testing.

@cedrict @tjhei @bobmyhill - The PR contains a working example (it runs).

The code can be cleaned up quite a bit (see commented ToDo), and one question to consider is whether we want to calculation the dilation rate from an average friction angle across compositions (current approach) or calculate a dilation rate for each composition and then average them (would require refactoring).

For some reason the dilation values are not being written out as a visualization field (they should be when selecting named additional outputs), but I probably just forgot to enable something.

Time to start testing :)

For all pull requests:

• [x] I have followed the instructions for indenting my code.

For new features/models or changes of existing features:

• [ ] I have tested my new feature locally to ensure it is correct.
• [x] I have created a testcase for the new feature/benchmark in the tests/ directory.
• [ ] I have added a changelog entry in the doc/modules/changes directory that will inform other users of my change.

[naliboff]

@tjhei - I've updated the PR to (1) remove the benchmark prm, (2) add a test case based off of the benchmark prm, (3) improve the code structure a bit where we calculate the dilation rate.

The convergence behavior is still poor for the cases with prescribed dilation turned on, but my recollections is that @cedrict also ran into these issues as well with his models? For ASPECT, a minimum resolution (5 refinement levels for the test case) is required to get to 1e-3 for standard picard iterations (defect correction does equally poorly).

The plot below shows the results for plastic dilation turned off (left column) versus turned on (right column) for different global refinement levels. The model with plastic dilation turned on and a refinement level of 4 has not yet converged. However, the plastic dilation is still clearly producing relatively resolution-independent shear band angles.

At this stage, it is an experimental feature that will require additional (lots?) of work for robust usage. Do we want to merge this features in a piecemeal fashion, or all at once when we are confident it works robustly?