Rewrite fragment evaluation to do less guessing

[stubailo]

TODO:

• If this PR is a new feature, reference an issue where a consensus about the design was reached (not necessary for small changes)
• Make sure all of the significant new logic is covered by tests
• Rebase your changes on master so that they can be merged easily
• Make sure all tests and linter rules pass
• Update CHANGELOG.md with your change
• Add your name and email to the AUTHORS file (optional)
• If this was a change that affects the external API, update the docs and post a link to the PR in the discussion

[stubailo]

I'm going to make a bold claim here: We should remove the returnPartialData option from writeQueryToStore and readQueryFromStore functions entirely.

I think it makes sense as an option to watchQuery - you don't want to get data before the loading state is false, but in other situations I can't imagine why it would be helpful to throw an internal apollo client error rather than returning data. Since the data is guaranteed to come back from the server pre-validated, the only situations this could be relevant are:

1. When validating the results of query reducers, which is a separate problem that we should tackle in a special way
2. When the store is updated partially to a state that the server would never return

Both of those seem only really solvable in the abstract with full knowledge of the schema, so having a half-baked way to identify errors by throwing on missing or extra fields doesn't seem like a good idea since it will almost certainly cause more problems (by being implemented incorrectly) than it solves.

Curious to get your thoughts on this, @helfer.

[stubailo]

OK I ended up making a lot of changes - @helfer ping me so we can talk before merging.