Use Shapeless 2.2.5 for 0.2.0

[travisbrown]

After publishing 0.1.2 this morning I realized that after some recent changes to generic and the upgrade to export-hook 1.0.2, we no longer absolutely need new changes in Shapeless 2.3.0.

The changes here would allow us to publish circe 0.2.0 today against cats 0.2.0 and Shapeless 2.2.5. We'd then be able to spend some time with the bigger changes in #78, #79, and #85 and publish them in circe 0.3.0 when Shapeless 2.3.0 and cats 0.3.0 are available.

I hadn't fixed #55 and #57 because I was expecting them to be resolved when Shapeless's #453 is closed, but if we're publishing 0.2.0 against Shapeless 2.2.5 it makes sense to fix them now (so I did). If we go this route I'll open an issue about removing the ambiguous instances before 0.3.0

[travisbrown]

For some reason the non-compilation tests that confirm that there are no Decoder[Object], etc. instances when using fully-automatic derivation are failing on 2.10 (using either ScalaTest's assertTypeError or Shapeless's illTyped), even though these instances aren't available either in the REPL or in a normally compiled program.

Because we're blocking these instances primarily to avoid confusing users, and since all other tests pass, I'm tempted to write this off as an issue with export-hook (or the non-compilation test mechanisms) and simply skip the tests on 2.10 (which is what I'm currently doing in this PR).

[travisbrown]

Okay, I'm going to fix #55 here but kick #57 down the road because I don't want to get bogged down in fighting with test-related issues that only show up on 2.10.

[codecov-io]

Current coverage is 69.56%

Merging #87 into master will increase coverage by +0.38% as of 5433731

Powered by Codecov. Updated on successful CI builds.

[ngbinh]

👍 would love to use circe now.

[yanana]

👍

[travisbrown]

This PR has become a kind of grab-bag of last little things for 0.2.0, but I think they're mostly uncontroversial, so once it's green I'll publish 0.2.0.