fix: add sdk:gradio vs sdk:docker warning — Dockerfile ignored in gradio mode

Agents kept writing Dockerfile fixes (pip uninstall, FROM python:3.X) that
had zero effect because README.md has sdk:gradio, which makes HF ignore
Dockerfile entirely. Added explicit warning to check sdk field first.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

scripts/conversation-loop.py

@@ -1120,6 +1120,7 @@ Priority: SURVIVAL first, then CAPABILITY. A dead child can't learn.
 - ⚠️ HF Spaces Docker SDK may OVERRIDE the base image Python version. Changing `FROM python:3.X` in Dockerfile does NOT guarantee that Python version runs. If a dependency fails due to Python version incompatibility (e.g. pydub needing pyaudioop removed in 3.13), the CORRECT fix is to REMOVE or REPLACE that dependency — NOT keep rewriting the Dockerfile.
 - If you've tried the same fix 3+ times and the error persists, CHANGE STRATEGY. Try removing the problematic dependency, using an alternative library, or wrapping the import in try/except.
 - If a removed dependency STILL appears in runtime errors, it is cached in Docker layers or installed as a transitive dep. Fix: add `RUN pip uninstall -y PACKAGE 2>/dev/null; true` AFTER `pip install` in Dockerfile. Also grep ALL code files for `import PACKAGE` and either remove or wrap in try/except.
+- ⚠️ CRITICAL: Check README.md `sdk:` field! If `sdk: gradio`, the Dockerfile is COMPLETELY IGNORED — HF uses its own Python environment. Dockerfile fixes (pip uninstall, FROM python:X) have NO effect. To make Dockerfile work, set `sdk: docker` in README.md. Alternatively, fix the issue in Python code (try/except imports).
 - NEVER install torch or transformers unless absolutely required — they are 2GB+ and cause OOM on free-tier Spaces. Use lightweight alternatives.
 
 MULTI-ACTION STRATEGY: