fix: unique_ids has exponential running time if no ID fields

[jpmckinney]

If instance has no ID fields, uniq(instance) isn't called once, but as many times as the length of instance.

Original commit d28dd01

[jpmckinney]

cc @duncandewhurst for prioritization

[jpmckinney]

Edit: I fixed the integration workflow.

---

You can ignore the ocds integration test failure. I assume it has something to do with how you install dependencies (it is not replicating how cove-ocds does it, which installs a single requirements file, whereas you install several which can cause different versions to be used).

        pip install -r cove-${{ matrix.cove }}/requirements.txt
        pip install -r cove-${{ matrix.cove }}/requirements_dev.txt
        pip install -r requirements_dev.txt
        # Make sure we're using local libs rather than one brought in
        # via requirements.
        pip install -e .
        pip install -e ./lib-cove-${{ matrix.cove }}/
        pip list

It's also possible you need to upgrade the workflow from Python 3.8. Our CoVEs use 3.10.

[jpmckinney]

For my own learning: This performance issue was hard to pinpoint, because the process never terminated (otherwise, we could use many typical profiling tools). I used py-spy's top -p command to monitor the PID of the Django process. I saw that jsonschema was spending a lot of time in equal, which looked like it was called from uniq. I then checked our code to see where uniq was called. (I also put various raise Exception to see which code was reached before the performance issue – an actual debugger would have probably been useful here.)

[jpmckinney]

Note to self to switch back to libcove release in cove-ocds once this is merged and released.