The current snapshot-based is getting unwieldy:

• Some of the fixtures are quite large and it's increasingly hard to understand what each test is verifying and what to expect from the output.
• It runs every test through every env, even though some of the tests' output is only useful to compare across a subset of the envs.
• Including the input code as one of the snapshots was well-intentioned, but, in practice, it's hard to parse the snapshot file well enough to directly compare input/output.
• GitHub's diff rendering of the snapshot file often fails to correspond snapshots based on their title, so the output is a mess.

I think we should try out inline snapshots with a new test suite that is more locally-scoped. We can include a few kitchen sink tests to get the high-level integration coverage that our current suite covers as well.