🛠️ Normalize tool-call IDs; harden quota redaction (#33)

* 🛠️ Normalize tool-call IDs; harden quota redaction (#30)

- Ensure `tool_calls[].id` / `tool_call_id` are deterministic and <= 40 chars (prevents OpenAI/LiteLLM 400s on GPT-5.2 mid-switch)
- Remove `/responses` `fc_` prefixing and use shared ID normalization
- Tighten quota/rate-limit detection to avoid false tool redactions; sanitize echoed context in logs
- Add/update unit tests + regression coverage for overlong tool IDs and redaction false-positives

* fix: Fixing broken build from redacting the crypto module.

Issue is based from trying to include crypto in both builds.  Which for the web build. causes the failure to occure.  Switched to a bit more of an in-house solution for stable hash which is used in dealing with tool ids.