juliet v.2.2.5

Fixed some bugs, in particular:

• Multi-instrument fits not updating posteriors by @Jayshil (PR #112).
• Multi-planet TTV fits did not work in previous version, now they do (issues #110 and #97) by @melissa-hobson (PR #111).
• Updated transit and eclipse fits to now remove fp component to out-of-eclipse component --- joint transit and eclipse models have thus, by default, and out-of-eclipse and out-of-transit value equal to 1.
• Setup file (and thus juliet) now installs h5py by default (needed for radvel) (PR #113).

This version also has new additions, namely:

• Now get_all_TESS_lightcurves has an extra flag to save lightcurves (save_data = True; PR #106 by @melissa-hobson).
• Added tests for transit, eclipse and joint transit and eclipse fits in tests folder.
• Implemented phase-curve toy model (simple sinusoid with phase-offset; amplitude set by secondary eclipse depth fp).
• Added phase-curve test suite under tests folder.
• juliet now supports light-travel time delay in eclipse and transit+eclipse fits (PR #113 by @taylorbell57). Activate by dataset.fit(..., light_travel_delay = True, stellar_radius = your_value). Note this applies the correction in this PR only on the eclipses via comparing radial distances from the eclipses to the time-of-transit center, which generates time delays which are subtracted iteratively to the measured time-stamps. This means the transits (or, really, the time-of-transit t0) are used as the references for the correction. In general practice this has little impact, except for (a) distortions that might be injected in phase-curves and (b) for comparisons with codes that apply this to the entire orbit (e.g., starry). On this latter ones, the reference to measure times of the orbit is typically at the center of mass of the system (or the star); in juliet, the reference for the time-stamps is the mid-transit point. As a practical example: for Earth, starry would measure a transit 8 mins earlier and an eclipse 8 minutes late. With juliet, you would measure the transit at t0, and the eclipse 16 minutes late to the no-delay case. The impact on the orbital parameters is the same in both cases.