Update to 0.6.2, defer pytorch variant selection to user

[bollwyvl]

Checklist

• Used a fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

Needed (maybe) for conda-forge/allennlp-feedstock#5

Adds tests. The tests download big models. They should probably be managed as additional sources and outputs. There is no good way (a la spacy) to cache the downloaded models at present. Really a bummer, they are quite large. Tests did pass locally on linux, though 😝

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I wanted to let you know that I linted all conda-recipes in your PR (recipe) and found some lint.

Here's what I've got...

For recipe:

• noarch packages can't have selectors. If the selectors are necessary, please remove noarch: python.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[bollwyvl]

(well, the tests worked locally except for the ones that needed the spacy en model, which i didn't add to the deps anyway)

[bollwyvl]

Tried replacing pytorch with pytorch-cpu, and lo, it works.

It seems like knowing how to installl CPU/GPU variants is a A Thing with torch, so expecting a user to have to say:

conda install -c conda-forge -c <somebody> pytorch-<whatever> pytorch-pretrained-bert-feedstock

is not entirely unreasonable, if somewhat unpleasant...