### Check for existing issues

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

### What happened?

## Summary

When the OpenTelemetry integration is enabled, `_otel_internal["spans_logged"]` stores **tuple keys** inside the request `metadata`. Because `litellm_pre_call_utils.py` shallow-copies `metadata` by reference into `proxy_server_request["body"]`, those tuple keys propagate into the spend-tracking path. `spend_tracking_utils.py` then calls `json.dumps` on the request body, which **crashes on every single request** with a `TypeError`.

The practical impact is catastrophic: every request fails to record its spend-log payload, so the in-memory `spend_log_transactions` queue grows without bound, leading to a severe memory leak and eventual OOM kills.

## Environment

- **LiteLLM version:** `1.85.0` - **Python version:** `3.13` - **OpenTelemetry integration:** Enabled (via `otel` callback) - **Proxy mode:** Yes

## Reproduction Steps

### 1. Minimal code trigger

```python import json from litellm.litellm_core_utils.safe_json_dumps import safe_dumps

# This is what opentelemetry.py injects into metadata via _otel_internal metadata = { …