Allow merge to work on VarInfo with different distributions

[torfjelde]

Expected behavior of merge is that it gives precedence to the right-most argument, but currently we raise an error in the scenario where the distributions between two varinfos don't match. This shouldn't be done (and is prohibitive to making the Gibbs sampler in TuringLang/Turing.jl#2099 work).

This PR fixes this.

[coveralls]

Pull Request Test Coverage Report for Build 6905019250

• 6 of 6 (100.0%) changed or added relevant lines in 1 file are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage decreased (-0.01%) to 80.247%

---

Files with Coverage Reduction	New Missed Lines	%
src/varinfo.jl	1	86.61%

Totals
Change from base Build 6895060737:	-0.01%
Covered Lines:	2535
Relevant Lines:	3159

---

💛 - Coveralls

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (485ebfb) 80.25% compared to head (94744df) 80.24%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #562      +/-   ##
==========================================
- Coverage   80.25%   80.24%   -0.02%     
==========================================
  Files          25       25              
  Lines        3161     3159       -2     
==========================================
- Hits         2537     2535       -2     
  Misses        624      624              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[yebai]

Thanks @torfjelde - it looks good to me!