Add HTTP API server, Android app, and CI build/publish pipeline

- docker-compose.yml + tooling-container/src/server.py: serve
  preprocess/OCR/extract over an async job API (POST /jobs, GET
  /jobs/{id}) instead of manual host-script runs per receipt
- android/: Kotlin/Compose app - capture a receipt photo, poll job
  progress, review/edit extracted fields, save locally (Room), export
  to CSV/ODS
- .gitea/workflows/build.yml: build the debug APK and publish it to
  dl.ladkau.de on push
- build-image.sh: also pull ollama/ollama so all pipeline images are
  ready after one script
- add project logo (also used for the app's launcher icon/splash)

tooling-container/src/server.py

@@ -0,0 +1,128 @@
+#!/usr/bin/env python3
+"""HTTP API that serves the pipeline (preprocess -> ocr -> structure) to
+the Android app. Runs the same in-process functions the numbered host
+scripts drive via `docker run` per stage, but skips the per-stage
+container start-up since a live server has no need for that
+inspectability - the host scripts remain the way to inspect each stage
+independently during pipeline development.
+
+Jobs are processed asynchronously so the app can show real per-stage
+progress: POST /jobs returns immediately with a job id; GET /jobs/{id}
+reports which of preprocessing/ocr/extracting is currently running, and
+the final result once done. Job state lives in memory only - this is a
+personal, LAN-only, single-instance server with no need for persistence
+here (the Android app is the system of record for confirmed receipts).
+"""
+import os
+import shutil
+import tempfile
+import threading
+import time
+import uuid
+
+import requests
+from fastapi import FastAPI, HTTPException, UploadFile
+from fastapi.responses import JSONResponse
+
+from extract import structure
+from ocr import ocr as run_ocr
+from preprocess import preprocess
+
+OLLAMA_BASE_URL = os.environ.get("OLLAMA_BASE_URL", "http://127.0.0.1:11434")
+OLLAMA_MODEL = os.environ.get("OLLAMA_MODEL", "qwen2.5:3b")
+
+app = FastAPI(title="fuelish")
+
+_jobs: dict[str, dict] = {}
+_jobs_lock = threading.Lock()
+
+
+def _set_job(job_id: str, **fields) -> None:
+    with _jobs_lock:
+        _jobs[job_id].update(fields)
+
+
+def _ensure_model_pulled() -> None:
+    """Best-effort: wait for Ollama to come up (compose starts `api` and
+    `ollama` together, with no ordering guarantee beyond container start)
+    and pull OLLAMA_MODEL if it isn't already present, so a bare `docker
+    compose up` is enough without a manual `ollama pull` step. Runs in a
+    background thread - jobs submitted before this finishes will simply
+    fail with a clear error from Ollama."""
+    tags = None
+    for _ in range(60):
+        try:
+            tags = requests.get(f"{OLLAMA_BASE_URL}/api/tags", timeout=5).json()
+            break
+        except requests.RequestException:
+            time.sleep(2)
+    if tags is None:
+        return
+
+    have = {m["name"] for m in tags.get("models", [])}
+    if OLLAMA_MODEL in have or any(m.startswith(OLLAMA_MODEL + ":") for m in have):
+        return
+    try:
+        requests.post(
+            f"{OLLAMA_BASE_URL}/api/pull",
+            json={"name": OLLAMA_MODEL},
+            timeout=1800,
+        )
+    except requests.RequestException:
+        pass
+
+
+@app.on_event("startup")
+def _on_startup() -> None:
+    threading.Thread(target=_ensure_model_pulled, daemon=True).start()
+
+
+@app.get("/health")
+def health() -> dict:
+    return {"status": "ok"}
+
+
+def _process_job(job_id: str, image_path: str, scratch_dir: str) -> None:
+    try:
+        _set_job(job_id, status="preprocessing")
+        final_path = preprocess(image_path, scratch_dir)
+
+        _set_job(job_id, status="ocr")
+        ocr_txt_path = run_ocr(final_path, scratch_dir)
+        with open(ocr_txt_path, "r", encoding="utf-8") as f:
+            ocr_text = f.read()
+
+        _set_job(job_id, status="extracting")
+        result = structure(ocr_text)
+
+        _set_job(job_id, status="done", result=result)
+    except Exception as exc:  # noqa: BLE001 - reported to the client, not swallowed
+        _set_job(job_id, status="error", error=str(exc))
+    finally:
+        shutil.rmtree(scratch_dir, ignore_errors=True)
+
+
+@app.post("/jobs", status_code=201)
+async def create_job(image: UploadFile) -> JSONResponse:
+    job_id = uuid.uuid4().hex
+    scratch_dir = tempfile.mkdtemp(prefix=f"fuelish-{job_id}-")
+    image_path = os.path.join(scratch_dir, "upload" + os.path.splitext(image.filename or "")[1])
+    with open(image_path, "wb") as f:
+        f.write(await image.read())
+
+    with _jobs_lock:
+        _jobs[job_id] = {"status": "queued", "result": None, "error": None, "_created": time.time()}
+
+    thread = threading.Thread(target=_process_job, args=(job_id, image_path, scratch_dir), daemon=True)
+    thread.start()
+
+    return JSONResponse({"job_id": job_id}, status_code=201)
+
+
+@app.get("/jobs/{job_id}")
+def get_job(job_id: str) -> dict:
+    with _jobs_lock:
+        job = _jobs.get(job_id)
+        if job is None:
+            raise HTTPException(status_code=404, detail="job not found")
+        return {"status": job["status"], "result": job["result"], "error": job["error"]}