Add gridded layout optimizer

[misi9170]

This pull requests adds a new "optimization" method that simply searches for the gridded layout (either a square grid or a hexagonal grid) that fits the most wind turbines into a specified boundary area. The spacing in the grid is fixed by the user, but the grid's rotation and translation is optimized (via a simple parameter sweep) to fit the most turbines into the boundary.

FlorisModel.run() is not called, so there is no wake or AEP calculation---the optimizer simply chooses the grid that fits the most turbines. Where multiple rotations/translations result in the same number of turbines, the result returned is (arbitrarily) the first of those. @ejsimley has suggested we could consider a "tie-breaker" of running an AEP calculation for all layouts that have the maximum number of turbines and selecting that with the highest AEP; for now, this has been left as future work.

There is no requirement that the boundary be a single area---users may specify a boundary with separate areas without issue.

The core code is complete. However, it currently runs somewhat slower than I'd hoped, and I intend to profile the code to see if there are any easy ways to speed it up.

To do:

• Core code implemented
• Existing tests pass
• New tests added
• Speed profiled
• Low-hanging fruit for speed-ups exploited
• Example demonstrating capability
• Documentation

[misi9170]

After profiling, not surprisingly it appears that the slowest part of the code is that which checks, point by point, whether candidate locations are inside the boundary. This code relies on shapely's contains method, which is called inside a nested loop: the outer loop is over the candidate rotations/translations, and the inner loop (represented as a list comprehension) is over the points within the candidate layout.

I tried a couple of quick things to speed this up, both of which failed:

1. Using numba's @jit decorator on the test_point_in_bounds method, which is simply a wrapper for shapely.contains() (numba could not recognize the shapely objects).
2. Running shapely.contains() on a MultiPoint object, rather than looping over each Point, to try to remove one level of the nested loop (shapely.contains() appears to check whether all of the points are contained in the shape and returns a scalar boolean value, rather than checking whether each point is contained as we need).

Overall, I think we can treat the code as fast enough for now. When I run the new example 004_generate_gridded_layout.py with min_dist_D=0.5, the 1000m x 1000m square boundary can fit 256 turbines, and the code spends about 1s in the test_point_in_bounds function running on my laptop.

[misi9170]

@paulf81 this is now ready for review

[paulf81]

Just curious why 1.1?

[misi9170]

Just as a little bit of over engineering to ensure that d is sufficient to cover the boundary area, even after translations/rotations. May not be necessary, but I thought just easy to add a safety factor.

[paulf81]

Overall this looks really great! All tests pass and examples run and produce clear results. Doc is also very clear. Approving and leave remaining minor comments up to you @misi9170