refactor: Optimise ID Generation

[lawrenceadams]

Currently there are ID generating methods at play in several models which are expensive requiring ordering of IDs (and hence lots of table scans); for little/no upside as not deterministic.

We should simplify and unify the approach across all models.

---

Performance on "big Synthea" continues to improve with #111 - at least for me! Feel free to test it yourselves by checking out my branch and giving it a run :)

btw @lawrenceadams I am thinking we might want to get rid of all the order by columns in our row_number ID assignments. it's a drag on performance with little benefit since it's rarely (if ever) actually deterministic. i think there's a whole separate discussion worth having on the best way to approach ID assignment (which can extend into topics like support for incremental runs). thoughts?

Originally posted by @katy-sadowski in #91