feat: Add memory system with SQLite + ChromaDB hybrid storage

- memory_store.py: User-isolated observation storage with vector embeddings
- New endpoints: /memory/save, /memory/query, /memory/get, /memory/timeline
- Progressive disclosure pattern for token-efficient retrieval
- Updated Dockerfile to ROCm 7.2 nightly

Dockerfile

@@ -0,0 +1,35 @@
+FROM python:3.12-slim
+
+# System deps: pdfplumber, ffmpeg for video audio extraction, build tools
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        build-essential \
+        curl \
+        ffmpeg \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+# Install PyTorch with ROCm support first (big layer, cache it)
+RUN pip install --no-cache-dir \
+    torch torchvision torchaudio \
+    --index-url https://download.pytorch.org/whl/nightly/rocm7.2/
+
+# Install remaining Python dependencies
+COPY app/requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy application code
+COPY app/ .
+
+# Pre-download the embedding model at build time so startup is fast
+RUN python -c "\
+from sentence_transformers import SentenceTransformer; \
+m = SentenceTransformer('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2'); \
+print('Model cached:', m.encode(['test']).shape)"
+
+EXPOSE 8899
+
+VOLUME ["/app/data", "/app/logs"]
+
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8899", "--log-level", "info"]