CV2-5640 only send responses to check api when we have a fully realized response from presto during sync requests

[DGaffney]

Description

Updates to prevent cases where we send multiple partial matches back to check-api before we've run all vectorizations. The pop that we are removing is theorized as the culprit due to several factors:

1. The pop removes the model, which, on response, makes it appear that there are fewer items waiting to be processed,
2. The pop is only on sync requests, while the nearly identical async requests are using get and appear to be working correctly.

Reference: CV2-5640

How has this been tested?

Not yet tested locally

Have you considered secure coding practices when writing this code?

None

[sentry-io]

🔍 Existing Issues For Review

Your pull request is modifying functions with the following pre-existing issues:

📄 File: app/main/lib/elastic_crud.py

Function	Unhandled Issue
get_blocked_presto_response	KeyError: 'openai-text-embedding-ada-002' api.sim... Event Count: 2

---

_{Did you find this useful? React with a 👍 or 👎}

[skyemeedan]

Is the problem that we are destructively modifying the 'obj' object when the "models" list is removed? It seems like it shouldn't make a difference since it isn't referenced later, but .get seems like better syntax.

[DGaffney]

when we send these requests out, with the pop in place, it "tells" the alegre callback that is expecting fewer models than it really should be - by enforcing the get we (a) bring it in line with async and (b) probably eliminate erroneous calls back to check-api when alegre mistakenly things work is complete for every in-process vector

[skyemeedan]

seems fine to me, but I don't really have full context. Since this is kind of core functionality for presto I think we should have test coverage for it if it is breaking?