### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

### Description of Issue

When these 2 criteria are met: 1. TPM rate limit on key / team set to a reasonable value, around 50k, and 2. A request is made to `/v1/responses` endpoint with a base64-encoded image in the request, with input tokens significantly less than the TPM limit

a 429 response will be returned to the user, with the error message: `Rate limit exceeded for api_key: <hashed-api-key>. Limit type: tokens. Current limit: 50000, Remaining: 50000. Limit resets at: 2026-07-11 01:40:13 UTC`.

Pod logs show: `parallel_request_limiter_v3.py:436 - TPM reservation estimate: input=633912, max_tokens=0 (explicit=False), total=633912`

When the TPM rate limit is removed or increased to a very high value, the request succeeds and the usage object in the corresponding spend log shows that there were only ~1k input tokens, which could not have burst the 50k TPM limit: ``` "usage_object": { "total_tokens": 1108, "prompt_tokens": 1049, "completion_tokens": 59, "prompt_tokens_details": { "text_tokens": null, "audio_t…