Given the current state of the package ecosystem, you frequently need to invest some time in a package that you want to use in a production setting, to the point of becoming familiar enough with the internals to submit PRs. Generally, this is how open source works. Projects becoming as mature and polished as some prominent packages in the Python and R ecosystems is more of an exception than the rule, and they also started out this way.
I find that this strategy can easily backfire. Committing to Julia at this point requires a certain mindset, not unlike flying the Millenium Falcon: it is designed to be blazingly fast, but things can break down at inconvenient moments and then you have to be able to repair them. This is perhaps more suited for research than production.