Refactor seeding steps

[lolmaus]

Seeding steps are a mess. They are inconsistent and hard to grasp.

Also, this should be extracted into a converter:

    const typePlural = pluralize(camelize(typeRaw));
    assert(`Collection ${typePlural} does not exist in Mirage`, server.db[typePlural]);

[simonihmig]

What do you have in mind as an alternative?

[lolmaus]

1. Make it obvious from the step text whether the step accepts a raw JSON input or a table.
2. Remove relationship lookup from tables and instead use productIds pseudo-attrs.
3. Provide single-line and multi-line versions of JSON-driven steps.
4. Less confusing documentation.

[lolmaus]

5. Remove the @id special format.

[lolmaus]

Previously, I've been trying to kill two birds with one stone: simplicity and readability. It was a limited success and I'm not happy with the quirks our seeding steps have.

Thus, I want to remove all the magic from the existing seeding steps, as described above, and focus on simplicity.

For readability, I have created a separate issue: #49. Set it to the farther milestone, since it's not a priority for us. Might make sense to remove it from the milestone.