fix: Handle multiple rate limit types per descriptor and prevent IndexError

[orolega]

Title

Fix IndexError in parallel request limiter v3 when handling multiple rate limit types per descriptor

Relevant issues

Fixes #15716

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🐛 Bug Fix
✅ Test

Changes

• Fixed IndexError in async_pre_call_hook by replacing fragile index mapping (floor(i / 2)) with a robust lookup using descriptor_key in rate limit statuses.
• Updated logic to match statuses to descriptors by key, supporting multiple rate limit types (requests, tokens, max_parallel_requests) per descriptor.
• Added fallback handling for cases where a status's descriptor_key does not match any descriptor, ensuring graceful error messages.
• Added/updated tests to cover scenarios with multiple rate limit types per descriptor and mismatched descriptor keys.

[vercel]

@orolega is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.