Deduplicate pricing lookup warnings for custom/unknown models

[pgjh]

Problem

When using Headroom proxy with a custom or unknown model (e.g. glm-5.2 routed through an OpenAI-compatible provider), LiteLLM cannot resolve pricing for that model. This triggers a WARNING log on every single request:

WARNING - headroom.proxy - Failed to get pricing for model glm-5.2: litellm.BadRequestError: LLM Provider NOT provided. Pass in the LLM provider you are trying to call. You passed model=glm-5.2

In an active session this floods proxy.log with hundreds of identical warnings — one per API call. While proxy.log has rotation, the log still grows rapidly and makes it harder to spot genuinely useful warnings.

Location

headroom/proxy/cost.py, line ~731:

except Exception as e:
    logger.warning(f"Failed to get pricing for model {model}: {e}")
    return None

Suggested Fix

Deduplicate the warning per model so it only fires once per unique model name per process lifetime:

except Exception as e:
    if not getattr(logger, "_warned_models", set()):
        logger._warned_models = set()
    if model not in logger._warned_models:
        logger._warned_models.add(model)
        logger.warning(f"Failed to get pricing for model {model}: {e}")
    return None

This is a minimal, non-invasive change — no new dependencies, no config flags, just a set on the logger object to track which models have already been warned about.

Environment

• Headroom version: 0.32.1
• Model: glm-5.2 via OpenAI-compatible backend (--backend anyllm --anyllm-provider openai)
• LiteLLM: installed (cannot resolve custom model names, as expected)

Impact

Low severity — does not affect functionality (pricing returns None, requests succeed). But the log spam is a real operational annoyance, especially for long-running proxy sessions.

[Parideboy]

Fixed in #2505. Dedupes the pricing-lookup warning per unresolvable model (module-level set, one warning per model per process lifetime) instead of logging on every request.