Add glance methods and IV diagnostics

[lukesonnet]

In response to #282, and for general compatibility, we are adding glance methods for our four classes of estimators (lm_robust, iv_robust, difference_in_means, horvitz_thompson).

This entails adding some simple design details (N, N_blocks, design_type) for some of the non-regression estimators, and adding some diagnostics for the regression estimators, most notably 2SLS IV (Wu-Hausman, first stage F test, Sargan).

This has all been implemented, the diagnostics have been tested against ivreg and stata when possible, and sanity checks have been placed on the glance methods.

Notably, we don't return IV diagnostics with fixed_effects, warning instead, and we silently return NA for the overid tests with weights, and this is documented in ?iv_robust.

Also see issue #287 for documenting this with gt and gtsummary.

[nfultz]

So far so good.

[lukesonnet]

Thanks for the comments Neal. Will implement tomorrow.

Note that in Slack we have decided not to return the first stage f-stats in glance.iv_robust if there are multiple endogenous regressors. See the problem posed here: tidymodels/broom#599

I will still compute the first-stage f-stats but they will only be visible in the returned object and when you print summary.iv_robust, not in glance.iv_robust.

[lukesonnet]

This PR is ready to be merged pending code review and tests passing.

[nfultz]

LGTM, very nice. See comments for minor tweaks.

[lukesonnet]

This is ready to be merged in. The only failure is on the branch travis, where the changes to the tests that are now in master have not been updated. You'll notice both PR tests have passed.

Please approve and we will merge.

[graemeblair]

LGTM. Amazing work.