I will test this now in practice to try to make a new build and will push fixes here. Until then, feel free to review/comment @cthoyt .

This is basically working now. One thing I didn't figure out how to refactor best is the integration of node assembly with cached node generation from individual processors. Currently, indra_cogex.representation.Node objects use INDRA-standardized namespace and ID pairs whereas the dumped tables use a single CURIE string with a different namespace and ID standard. Assembly works directly on Node objects which means that assembly can only happen if nodes are collected from processors. In principle we could re-read the dumped node tables into Node objects, then assemble them, and dump them again into an assembled table but that seems a bit "dirty" to me (instead of just working with the original Node objects that processors generate).

Overall looks good. I think we should have started by keeping the prefix/id separate for each node. I made a comment on the code, but the operation of validating prefix/id pairs and canonicalizing them should be operations implemented in INDRA, right?

Also, now we have the nice CI tests for code docs, we have to do all of them ;)

The tests are erroring on GHA with ModuleNotFoundError: No module named 'indra_cogex', with this setup I'm not sure what to change for the package to be visible.

I'm honestly not sure this much strict linting and mypy is productive, we need to be able to move on and focus on functionality.

The tests are erroring on GHA with ModuleNotFoundError: No module named 'indra_cogex', with this setup I'm not sure what to change for the package to be visible.

@cthoyt I'm not sure why this happens, I haven't used tox with pytest in this way before so am a bit lost. What would we need to change to avoid the "ModuleNotFoundError: No module named 'indra_cogex'" errors on GHA? See e.g., https://github.com/bgyori/indra_cogex/runs/2593007487?check_suite_focus=true.