add nm_join_bayes

[kyleam]

When working on run_sims in gh-75, we decided to focus on the return value and rename the function to nm_join_bayes, treating it as the nmbayes analog of nm_join. Going that direction led to gh-96, and consensus from discussion there was that 1) we should try to align with nm_join as closely as possible (e.g., support .files) and 2) ideally nm_join would be used underneath to avoid repeating some its more complicated logic.

This series implements that proposal and supersedes gh-75. Most notable changes from gh-75:

• Use nm_join underneath.

• Don't write behind-the-scenes table when submitting model; instead use the nm_join default for the tables and allow callers to specify file via .files.

• Drop argument for specifying custom data; instead add an argument that allows callers to specify a function that can be used to transform the data after the table-to-data join but before simulating with mrgsolve.

• Support .superset = TRUE.

• Drop log_dv argument; see discussion in gh-97. (I'll follow up there.)

• Flesh out tests. At some point, @seth127 and I think it'd be nice to have the SME-inspected tests along the lines of what's discussed in gh-75, but we're not sure what that will look like, and it's likely to live outside this test suite, so we're not considering it a blocker for merging this PR.

Closes #46. Closes #50. Closes #96. Closes #100.

Re: #97

cc: @kylebaron @timwaterhouse

---

• [x] Open PR nm_join "origin attribute" PR for bbr. (This PR introduces handling for that feature but also has fallback code for older bbr versions.) update: https://github.com/metrumresearchgroup/bbr/pull/617

[kyleam]

@seth127 Thanks for the review. Pushed updates. Back to you.