build(docker): slim backend image, static frontend image, -docling variant

Backend (arc53/docsgpt): 4.5 GB compressed -> 0.9 GB with both embedding
models and tiktoken baked in.
- torch/transformers gone from the default install (docling extra only).
- Ubuntu 24.04 ships python3.12: no deadsnakes PPA, no software-properties-
  common; every pin is a wheel, so no gcc/g++/rust in the builder.
- COPY --chown and a prefetch that runs as the process user replace the
  trailing chown -R, which duplicated the 600 MB model layer.
- .dockerignore keeps __pycache__, .coverage, local indexes and .env out.
- EXTRAS build arg (INSTALL_DOCLING kept as an alias); the docling variant
  also bakes docling's layout/table/RapidOCR models (DOCLING_ARTIFACTS_PATH)
  and tesseract, and drops only the discovery documents of Google APIs the
  app never builds.
- FLASK_DEBUG env removed (unused); OCI labels added.

Frontend (arc53/docsgpt-fe): 302 MB Vite dev server -> 25 MB static build
behind nginx. VITE_* variables are injected at container start into
/config.js and read through src/env.ts, so the image no longer needs a
rebuild per deployment; docker-compose.yaml keeps hot reload via the dev
target.

Publishing: every release and develop build now pushes a slim tag and a
-docling tag (docling engine + models + tesseract). docker-compose-hub.yaml
takes DOCSGPT_IMAGE_TAG / DOCSGPT_IMAGE_VARIANT; docker-compose-standalone.yaml
runs the stack from pre-built images without a checkout and is attached to
each release. setup.sh selects the -docling variant for OCR instead of
requiring a local build. A new workflow builds the image on PRs that touch
it and runs verify_offline under --network none; lint checks the exported
requirements match uv.lock.

setup.sh

@@ -367,29 +367,32 @@ configure_doc_processing() {
         echo -e "${GREEN}PDF-as-image parsing enabled.${NC}"
     fi
 
-    # OCR needs the tesseract binary, an optional system package that only
-    # locally built images can include (INSTALL_TESSERACT build arg). The
-    # pre-built Docker Hub images ship without it, so there OCR stays off
-    # (its default) rather than being switched on to fail on every scan.
-    if [[ "$COMPOSE_FILE" != "$COMPOSE_FILE_LOCAL" ]]; then
-        echo -e "${YELLOW}OCR for scanned PDFs and images stays off: the pre-built Docker Hub images do not include tesseract. To use OCR, rerun setup and choose option 5 (build images locally), or use a DeepSeek-OCR endpoint by adding OCR_ENABLED=true, OCR_ENGINE=deepseek and OCR_DEEPSEEK_URL=<endpoint> to .env.${NC}"
+    # OCR needs the tesseract binary. The default (slim) images ship without
+    # it; the pre-built "-docling" image variant bakes tesseract, the docling
+    # layout engine and its models in, so with Docker Hub images OCR means
+    # switching the variant. Locally built images get it via build args.
+    read -p "$(echo -e "${DEFAULT_FG}Enable OCR for scanned PDFs and images? (y/N): ${NC}")" ocr_enabled
+    if [[ ! "$ocr_enabled" =~ ^[yY]$ ]]; then
         return
     fi
-
-    read -p "$(echo -e "${DEFAULT_FG}Enable OCR for scanned PDFs and images? (y/N): ${NC}")" ocr_enabled
-    if [[ "$ocr_enabled" =~ ^[yY]$ ]]; then
-        echo "OCR_ENABLED=true" >> "$ENV_FILE"
-        # Bakes tesseract into the locally built images (docker compose
-        # --env-file .env build).
-        echo "INSTALL_TESSERACT=true" >> "$ENV_FILE"
-        echo -e "${GREEN}OCR enabled. tesseract will be built into the images (INSTALL_TESSERACT=true); for a DeepSeek-OCR endpoint instead, set OCR_ENGINE=deepseek and OCR_DEEPSEEK_URL=<endpoint> in .env.${NC}"
-        read -p "$(echo -e "${DEFAULT_FG}Also install the Docling layout engine for OCR (better tables/reading order, several GB heavier)? (y/N): ${NC}")" docling_ocr
-        if [[ "$docling_ocr" =~ ^[yY]$ ]]; then
-            # Locally built images include docling via this build arg; it becomes
-            # the OCR backend automatically (OCR_BACKEND=auto).
-            echo "INSTALL_DOCLING=true" >> "$ENV_FILE"
-            echo -e "${GREEN}Docling will be built into locally built images (docker compose --env-file .env build). Pre-built Docker Hub images do not include it.${NC}"
-        fi
+    echo "OCR_ENABLED=true" >> "$ENV_FILE"
+    if [[ "$COMPOSE_FILE" != "$COMPOSE_FILE_LOCAL" ]]; then
+        # Pre-built images: pull arc53/docsgpt:<tag>-docling instead of the
+        # slim default (about 1.5 GB more to download).
+        echo "DOCSGPT_IMAGE_VARIANT=-docling" >> "$ENV_FILE"
+        echo -e "${GREEN}OCR enabled. The -docling image variant will be pulled (tesseract, docling layout engine and its models included). For a DeepSeek-OCR endpoint instead, set OCR_ENGINE=deepseek and OCR_DEEPSEEK_URL=<endpoint> in .env.${NC}"
+        return
+    fi
+    # Bakes tesseract into the locally built images (docker compose
+    # --env-file .env build).
+    echo "INSTALL_TESSERACT=true" >> "$ENV_FILE"
+    echo -e "${GREEN}OCR enabled. tesseract will be built into the images (INSTALL_TESSERACT=true); for a DeepSeek-OCR endpoint instead, set OCR_ENGINE=deepseek and OCR_DEEPSEEK_URL=<endpoint> in .env.${NC}"
+    read -p "$(echo -e "${DEFAULT_FG}Also install the Docling layout engine for OCR (better tables/reading order, several GB heavier)? (y/N): ${NC}")" docling_ocr
+    if [[ "$docling_ocr" =~ ^[yY]$ ]]; then
+        # Locally built images include docling via this build arg; it becomes
+        # the OCR backend automatically (OCR_BACKEND=auto).
+        echo "INSTALL_DOCLING=true" >> "$ENV_FILE"
+        echo -e "${GREEN}Docling will be built into locally built images (docker compose --env-file .env build).${NC}"
     fi
 }