Revisit handling of non-semver package versions

[jonzeper]

We're filtering out non-semver versions from the version select, but someone could still add a non-semver versioned package in the YAML editor, and that causes a crash when we try to compare it.

• Let's not filter out the non-semver versions. Instead let's keep them but coerce them into valid semvers for the purpose of sorting.

• When determining the selected value, we can just use a simple string comparison, no need for special semver check

[anirrudh]

I'm going to fix the ESLint error here, but this makes sense to me, and I'll add it to the documentation - having the ability to coerce and manually override is a better strategy!

[jonzeper]

Thanks! I got a little confused what the linter's beef was there

[anirrudh]

Hmmm....adding semvers to the yaml causes the build to fail; tried with numpy and pyarrow (0.5.0.pre), is this behavior expected?

[jonzeper]

Not expected. I didn't test extensively, but was able to build an env with pyarrow>=0.5.0pre. Were you able to see a specific error?

[anirrudh]

Nope, just that the env didn't build. But the tests do, and you were able to install! I'm merging - perhaps it's because I did an == instead of >=, just know that mamba fails the build - most likely since it's pretty old. Something to take note of - otherwise, LGTM, merging!