Make tweaks for CRAN

[lukesonnet]

There were some "deep" CRAN errors we needed to resolve that you can see here: https://cran.r-project.org/web/checks/check_results_estimatr.html

1. Tests failed without Long Doubles (https://www.stats.ox.ac.uk/pub/bdr/noLD/estimatr.out). They were not identical but rather equal, so I just changed the test.
2. Uninitialized values (the degrees of freedom, residuals, and residual variance in lm_robust_helper.cpp were being used and returned to the R wrapper code and were causing errors in Valgrind. I now initialize these values and tested all of the test files one by one using valgrind and there are no more errors.
3. Solaris errored on a test that expect_equal the max diff of two margins was less than < 2e05. I changed the test to expect_equals with a tolerance of 0.01. I have no way to test this since rhub's Solaris platform apparently cannot compile RcppEigen files (and I'm not sure if I can fix this). See rhub Solaris fails to compile RcppEigen package r-hub/rhub#109.

I think if this passes, we might as well try to resubmit and see what they say?

Oh, I also got rid of my silly way to handle if the cholesky decomposition failed that relied on catching my own error. Logic should be much cleaner now.

[coveralls]

Coverage increased (+0.0002%) to 99.633% when pulling a7c10c8 on lukesonnet/cran-iss into ad8f606 on master.

[lukesonnet]

Sorry @nfultz , I'm not sure where a lot of the diffs in the cpp came from, most are useless.

[nfultz]

Thank you adding do_qr - flow control via exceptions is unfun.

[nfultz]

LGTM, feel free to merge