Refactor: Infer source

[nikbpetrov]

Before merging, remember to remove -new suffix from the name of the new jobs in src/orchestration/func_infer_fetch/Makefile and src/orchestration/func_infer_update/Makefile

[houtanb]

First pass, quick review from codex and claude:

1. Avoid importing fetch/update dependencies at module import time — src/sources/infer.py:9-16 Moving the old INFER fetch/update code into sources.infer makes every consumer of the source registry import backoff, certifi, and requests just to load InferSource. That breaks entrypoints that import sources or helpers.question_curation but do not install those packages (for example src/orchestration/func_resolve/requirements.txt, src/leaderboard/requirements.txt, src/curate_questions/create_question_set/requirements.txt, and the metadata jobs), so those jobs will now fail during startup with ModuleNotFoundError before any work begins.
2. source_intro/resolution_criteria are ClassVars only on InferSource, no base class contract
3. orchestration/__init__.py docstring still says "resolve pipeline" but now also does fetch/update

Me quick review:

1. Redundant: source_type: ClassVar[SourceType] = SourceType.MARKET
2. What is display_name for? This should be "Rand Forecasting Institute", but I don't think we're ever displaying sources anywhere. Continue to use "infer" for historical reasons for "source"
3. On cursory review, it seems like now you're downloading all resolution files from GCP Storage whereas I think you only want those in dff, otherwise you'll be downloading all resolved resolution files as well, which will grow with time. Likely not a big deal for infer, but costly for metaculus and polymarket.

[nikbpetrov]

1. In-freaking-excusable. I get it's part consequence of the intermedia code structure, but still... all this testing and seemingly sensible arguments I put forward that a single test was going to be sufficient and it all proves to be little more than a fart in the wind. I will revert to running much more comprehensive test suite during this refactor moving forward and will init a separate repo where testing files will go. Will work on a fix too - not sure what the best option is.
2. Fixed, though strict enforcement is deferred (i.e. tests + typing check) until all sources are refactored as otherwise sources that are yet to be refactored will be throwing errors (we can add those vars there but that means touching many more sources prematurely)
3. Fixed
4. Fixed
5. Was meant to be for logging/debugging in my head based on the initial plan (when I wrote it), but you're right it's not used so I've removed
6. Good catch; I've added an optional arg to pass ids so we don't load everything. The only hypothetical caveat is that files whose ids are not passed, i.e. not in dff, won't get updated - safe assumption, right?

[houtanb]

I don't see the new commits.

[nikbpetrov]

Yup, still working on it; don't have 1. solved yet, see Slack. Don't have the testing pipeline ready yet either, not continuing without one at this point.

[houtanb]

Of course, sorry, did not mean to pressure you. I saw "fixed" for some of them so was looking for those changes. Take your time, I'll respond to your Slack message tonight or tomorrow. Need to think a bit about it

[nikbpetrov]

Source identity split solved. Please note that I've touched a number of files almost exclusively due to the fact that I wanted to move resolution_criteria and source_intro into sources/_metadata.py. I have then used the import/reexport pattern in each helpers/<source>.py. Technically a handful of these files could be deleted altogether if I just changed the import pattern in question_curation to import from sources metadata, but I've refrained and will do that during the dedicated question_curation refactor PR - let me know if you prefer otherwise (e.g. so we don't forget later, but we should be fine!)