Allowing litellm<1.72 in llm extra for working VCR

[jamesbraza]

The backstory here is:

1. Pinned latest litellm for tool_choice=Tool's finish_reason fix #244 pinned litellm>=1.72.0 for improved finish_reason logic
2. However, last night I discovered [Bug]: #11097 broke HTTP request caching via vcrpy BerriAI/litellm#11724

So this PR:

• Restores the versions from Pinned latest litellm for tool_choice=Tool's finish_reason fix #244
  • With the exception of the paper-qa edge case (as thanks to Supporting Python 3.13 paper-qa#946, paper-qa now supports Python 3.13)
• Adds logic to check the litellm version at runtime, auto-selecting the appropriate validation behavior
• Refreshes all VCR cassettes using ToolSelector

[Copilot]

Pull Request Overview

This PR adjusts the version pinning for litellm in the llm extra to allow versions below 1.72 so that VCR cassettes won’t break, and it adds a runtime version check to automatically select the correct finish_reason validation logic.

• Restores a lower version constraint for litellm in the llm extra while adding an import and runtime version check in utils.py
• Updates cassette fixtures (headers, content-length, user-agent, etc.) to reflect new version information
• Revises pyproject.toml to tighten the allowed litellm version range and add a packaging dependency

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/cassettes/TestParallelism.test_tool_selector_with_external_acompletion*.yaml	Updates request/response formatting and header values for consistency and versioning
tests/cassettes/TestParallelism.test_tool_selector_from_model_name*.yaml	Similar cassette updates with differences in header values (e.g. read timeout)
tests/cassettes/TestParallelism.test_dummyenv_using_empty_params.yaml	Adjusts JSON payload formatting and header updates
src/aviary/tools/utils.py	Adds a runtime version check for litellm and introduces a new flag for finish_reason logic
pyproject.toml	Changes litellm version constraints and adds packaging as a dependency

Comments suppressed due to low confidence (1)

tests/cassettes/TestParallelism.test_tool_selector_with_external_acompletion[gpt-4o-mini-2024-07-18].yaml:39

• The header 'x-stainless-read-timeout' is set to '6000.0' in this cassette while a similar cassette uses '600.0'. Please confirm that this difference is intentional to avoid potential inconsistency issues.

x-stainless-read-timeout:

[maykcaldas]

Nice dive 🙌🏻