Develop merge of #671

[rileyjmurray]

The auto-merge of #671 from bugfix into develop failed, so this PR is handling that manually.

Description from bugfix merge below.

This PR is split off from #664. It fixes how wildcard error is computed in leakage modeling. When merged it will resolve #652. It also includes changes to the default leakage-aware gauge optimization suite. The latter changes are needed to prevent gauge optimization from trying to spuriously "spread out" relational errors across available gates.

This PR adds a meaningful correctness test for leakage GST modeling. To reduce the time of the test I used a huge number of shots per circuit in the simulated dataset. This led to errors with "min_prob_clip" constants in objective function evaluation. In order to resolve those errors I replaced hard-coded instances of that constant's default value (1e-4) to values of a module-wide constant; I made similar changes for "radius" constants in objective functions.

Finally, this PR adds implementations of frobeniusdist, frobeniusdist_squared, and residual to ModelMember, and removes the implementation of these functions from all child classes. Removing the child class implementation become possible because the ModelMember.residual implementation relies on a new method, ModelMember._to_transformed_dense(...), that child classes must implement. The semantics of _to_transformed_dense are documented in ModelMember. I provided implementations for base classes for POVM effects, states, and linear operators.

(This PR is a superset of the since-closed PR #670, which started from develop instead of bugfix.)