Sorting LiteLLM GitHub issues by difficulty to implement. Higher = harder.

{
#35524 requires designing a reliable atomic reservation fallback across unknown-cost and multimodal requests, preserving accounting correctness under concurrency and adding broad regression coverage. #31260 is primarily a scoped propagation and cache-metadata consistency fix across a few synchronous paths.
}
https://github.com/berriai/litellm/issues/35524 3:2 https://github.com/berriai/litellm/issues/31260