Support benchmarks

[leo-mazzone]

Context

Changes proposed in this pull request

• Generalised generation of dummy probabilities to support dedupers. This now uses deterministic logic for base edges, and generates extra edges by choosing among all possible edges (we need to avoid duplicate edges)
• Moved factories.py out of test, into src/common (because it's now used by the Postgres server benchmarks)
• Created sub-module for benchmarking Postgres
• Added script that outputs schema initialisation SQL
• Added script that generates all 6 Postgres tables
• Moved test_transform.py under common tests
• Solved quite a few bugs
• Added various tests

Guidance to review

• The logic for generating probabilities has changed such that we can use the output to produce tables conforming to the ORM constraints, as well as working with the logic for generating clusters. We need to write checks for the input to the cluster generation: no duplicate edges; no self-references; and for linking, ensure graph is bipartite? When do we do it?

Checklist:

• My code follows the style guidelines of this project
• New and existing unit tests pass locally with my changes

[wpfl-dbt]

Looking good.

Is it worth using mb.query() or mb.match() to test that the dummy data behaves as we expect?

[wpfl-dbt]

Approved pending unit tests passing.

[wpfl-dbt]

Happy with new changes