### Check for existing issues

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

### What happened?

Every `print_verbose` helper in the codebase prints its argument **raw** to stdout when `litellm.set_verbose is True`. Callers routinely pass dict-shaped diagnostic payloads (e.g. `model_call_details`) that contain `api_key` (and, on the proxy, master/virtual keys, and provider Bearer tokens), so the full secret is emitted verbatim to stdout and captured by container/pod log collectors. In addition, model- and request-controlled content reaches these same stdout paths (e.g. the async streaming line logger), so unescaped CR/LF/Unicode-separator/control characters allow log forging and terminal-escape injection (CWE-117).

The `verbose_logger` path is already redacted by `SecretRedactionFilter`; only the `set_verbose` stdout `print(...)` branch is unprotected. There are **multiple** copies of `print_verbose` across the code, and they share this same unredacted-stdout pattern:

- `litellm/utils.py::print_verbose` - `litellm/_logging.py::print_verbose` - `litellm/main.py::print_verbose` (prints provider streaming chunks — `line in async streaming`)…