Ensure virtual-key model budgets force router fallbacks

[mxrcooo]

Title

Ensure virtual-key model budgets force router fallbacks

Relevant issues

Fixes #15223

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

• Refactored async_log_success_event() to parse the real user_api_key_model_max_budget, resolve aliases, and record spend against the same cache keys that enforcement uses.
• Implemented async_filter_deployments() to mirror the deployment-level limiter flow:
  1. Read the virtual key’s budget from request metadata.
  2. Resolve aliases (e.g., model_name → litellm_params.model) and check cached spend for each deployment.
  3. Drop deployments whose spend exceeds max_budget.
  4. Surface a BudgetExceededError only after every deployment has been filtered, matching existing auth behaviour.
• Added unit tests to confirm spend tracking still works and that routing now excludes over-budget deployments (and raises when all fallbacks are exhausted).

Screenshot of passing tests + all make test-unit passes:

[vercel]

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

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.