feat(accent): local UniDic + POS-driven patches

.gitignore

@@ -216,4 +216,11 @@ __marimo__/
 .streamlit/secrets.toml
 
 # Secrest files
-secret*
+secret*
+
+data/
+output/
+
+CLAUDE.md
+AGENTS.md
+.claude

Dockerfile

@@ -2,20 +2,35 @@ FROM python:3.11-slim AS builder
 
 WORKDIR /app
 
-# Install uv for fast dependency management (pinned for reproducible builds)
-RUN pip install --no-cache-dir "uv==0.9.0"
+# Install uv for fast dependency management (pinned for reproducible builds),
+# curl for downloading the NINJAL UniDic archive (scripts/download_unidic.sh
+# line ~74), and unzip for extracting it (Deflate64-compressed).
+RUN pip install --no-cache-dir "uv==0.9.0" \
+    && apt-get update \
+    && apt-get install -y --no-install-recommends curl unzip \
+    && rm -rf /var/lib/apt/lists/*
 
 # Copy dependency files first for better layer caching
 COPY pyproject.toml uv.lock ./
 
 # Install dependencies using uv
 RUN uv sync --frozen --no-dev --no-install-project
 
+# Download and install the UniDic dictionary into the venv so the image is
+# self-contained — the `unidic` pip package ships the loader but not the
+# dicdir, and fugashi.Tagger() fails at runtime without it.
+# Uses scripts/download_unidic.sh so local dev and Docker share the same
+# download logic.  Kept as its own layer (before COPY . .) so app-code
+# edits don't bust the download cache; only re-runs when the script or
+# uv.lock changes.
+COPY scripts/download_unidic.sh scripts/
+RUN VIRTUAL_ENV=/app/.venv bash scripts/download_unidic.sh
+
 COPY . .
 
 # Compile only our app code (skip .venv) and drop the .py sources
-RUN python -m compileall -b -q main.py api config \
-    && find main.py api config -name "*.py" -delete
+RUN python -m compileall -b -q main.py api \
+    && find main.py api -name "*.py" -delete
 
 FROM python:3.11-slim
 
@@ -30,7 +45,6 @@ WORKDIR /app
 COPY --from=builder /app/.venv /app/.venv
 COPY --from=builder /app/main.pyc /app/main.pyc
 COPY --from=builder /app/api /app/api
-COPY --from=builder /app/config /app/config
 
 # Set PATH to use venv binaries
 ENV PATH="/app/.venv/bin:$PATH"