Fix underdetermined regularized least squares

[RobbieRao]

Summary

• build the Tikhonov regularization block from the number of coefficients
• append the matching number of zero rows to vector and matrix right-hand sides
• cover underdetermined systems in both solver implementations

Root cause

The augmented least-squares system used an N x M identity matrix and padded
the right-hand side with N rows, where N is the number of equations and M
is the number of coefficients. For underdetermined systems (N < M), this left
some coefficients unregularized and produced a solution that did not minimize
the documented ridge objective.

For example, with A = [[1, 1]], b = [1], and regularization weight 1, the
expected ridge solution is [1/3, 1/3]; the previous implementation returned
approximately [0, 1].

Impact

Underdetermined systems now solve the documented regularized least-squares
objective. Square and overdetermined systems retain the same mathematical
result.

Testing

• python gnm/shape/run_all_tests.py — 278 passed, 2 skipped
• pylint gnm/shape/fitting_utils/regularized_least_squares.py gnm/shape/fitting_utils/regularized_least_squares_test.py

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[RobbieRao]

@googlebot rescan