refactor: multi-provider memory adapters with scan-based builtin (#227)

* refactor: restructure memory into multi-provider adapters, remove manifest.json dependency

- Rename internal/memory/provider to internal/memory/adapters with per-provider subdirectories (builtin, mem0, openviking)
- Replace manifest.json-based delete/update with scan-based index from daily files
- Add mem0 and openviking provider adapters with HTTP client, chat hooks, MCP tools, and CRUD
- Wire provider lifecycle into registry (auto-instantiate on create, evict on update/delete)
- Split docker-compose into base stack + optional overlays (qdrant, browser, mem0, openviking)
- Update admin UI to support dynamic provider config schema rendering

* chore(lint): fix all golangci-lint issues for clean CI

* refactor(docker): replace compose overlay files with profiles

* feat(memory): add built-in memory multi modes

* fix(ci): golangci lint

* feat(memory): edit built-in memory sparse design

docker/Dockerfile.sparse

@@ -0,0 +1,43 @@
+# syntax=docker/dockerfile:1
+
+# Sparse encoding service — runs the OpenSearch neural sparse model
+# as a standalone HTTP service (Flask + PyTorch CPU).
+# Used by the Go server's sparse memory runtime via HTTP.
+
+FROM python:3.12-slim
+
+WORKDIR /app
+
+COPY internal/memory/sparse/service/requirements.txt requirements.txt
+RUN pip install --no-cache-dir \
+    --extra-index-url https://download.pytorch.org/whl/cpu \
+    -r requirements.txt
+
+COPY internal/memory/sparse/service/main.py main.py
+
+ENV SPARSE_PORT=8085
+ENV SPARSE_CACHE_DIR=/opt/sparse-cache
+
+RUN mkdir -p /opt/sparse-cache
+
+# Pre-download the default sparse model during image build so containers
+# start with a warm cache and do not need to fetch weights on first boot.
+RUN python - <<'PY'
+from pathlib import Path
+from huggingface_hub import hf_hub_download
+from transformers import AutoModelForMaskedLM, AutoTokenizer
+
+model_repo = "opensearch-project/opensearch-neural-sparse-encoding-multilingual-v1"
+cache_dir = "/opt/sparse-cache"
+Path(cache_dir).mkdir(parents=True, exist_ok=True)
+
+AutoModelForMaskedLM.from_pretrained(model_repo, cache_dir=cache_dir)
+AutoTokenizer.from_pretrained(model_repo, cache_dir=cache_dir)
+hf_hub_download(repo_id=model_repo, filename="idf.json", cache_dir=cache_dir)
+
+print(f"Pre-downloaded sparse model: {model_repo}")
+PY
+
+EXPOSE 8085
+
+CMD ["python", "main.py"]

docker/docker-compose.yml

@@ -31,6 +31,11 @@ services:
         - VITE_API_URL=${VITE_API_URL:-/api}
         - VITE_AGENT_URL=${VITE_AGENT_URL:-/agent}
 
+  sparse:
+    build:
+      context: .
+      dockerfile: docker/Dockerfile.sparse
+
   browser:
     build:
       context: .