Overall idea is: Core and Database contain base stuff and don't import
API. And API imports from Core and Database.

It also improves the error in case it's the first time we are running
the tests and we have no coreNLP service running.

This commit modifies the backend code to convert the input CSV into an
`NgramsList`, so that we can reuse the JSON API for the upload.

it also brings the CSV and JSON APIs closer in look.
Temporarily disable the `importCsvFile` function.

This commit makes possible to pass an optional `X-Garg-Error-Scheme`
header in a client request to enable the new error format.

Added more tests for more subtle cases.

Also, some fixes to NodeStory (it inserts ngrams when saving node
story, when they don't exist).

This commit introduces a new DB function called `isSharedWith`
that can be used to check for team/public/private nodes and their
shared resources.

This is a first stepping stone in avoiding bugs by having `UserId` and
`ContextId` being proper types, and not synonyms.

We also refactor things to have `AuthenticatedUser` carry the `UserId`.

This commit partially undo the access policy checks for the
GraphQL endpoints, until we find a better story to correlate
the `user_id` we sometimes receive with the info from the
`AuthenticatedUser`.

This commit makes the `GET /api/v1.0/version` test more lax, and less
prone to break on version bumps.

It can be used to "bolt-on" policy checking into Servant API routes.

Tests were written with a very flat module hierarchy which made them
a bit confusing to navigate. This commit moves everything under a common
`Test.*` hierarchy.