Standalone Shonar Desktop: vendor portable sources + local engine; decouple from ~/Projects/Shonar

- shared/ = portable Android-origin sources vendored from deferred/desktop-server
  (app/build.gradle.kts srcDir repointed; PlaybackController.kt excluded as Android-only)
- backend/ = bundled-lite engine (SQLite + inline queue); .venv symlinked from the
  old checkout, PYTHONPATH pins THIS backend's code over any editable install
- repoRoot() resolves this project dir (env SHONAR_REPO still wins); desktop-dev.sh
  watches shared/ + backend/
- Verified: :app:compileKotlin + :app:test green (23 tests); engine boots on :8010,
  self-migrates, /healthz ok

backend/shonar/services/ai/ollama.py

@@ -0,0 +1,78 @@
+"""Summaries via a local Ollama server (`/api/chat`, JSON mode)."""
+
+from __future__ import annotations
+
+from collections.abc import AsyncIterator
+from contextlib import asynccontextmanager
+
+import httpx
+
+from shonar.services.ai import ProviderConfigError, ProviderTransientError, SummaryResult
+from shonar.services.ai._llm import SYSTEM_PROMPT, build_user_message, parse_summary
+
+
+class OllamaProvider:
+    name = "ollama"
+
+    def __init__(
+        self,
+        base_url: str,
+        model: str = "",
+        timeout_s: float = 900.0,
+        http_client: httpx.AsyncClient | None = None,
+    ) -> None:
+        if not base_url.strip():
+            raise ProviderConfigError("ollama needs SHONAR_LLM_BASE_URL.")
+        if not model.strip():
+            raise ProviderConfigError("ollama needs SHONAR_LLM_MODEL.")
+        self.base_url = base_url.rstrip("/")
+        self.model = model
+        self.timeout_s = timeout_s
+        self.http_client = http_client
+
+    @asynccontextmanager
+    async def _client(self) -> AsyncIterator[httpx.AsyncClient]:
+        if self.http_client is not None:
+            yield self.http_client
+        else:
+            async with httpx.AsyncClient(timeout=self.timeout_s) as client:
+                yield client
+
+    async def summarize(self, transcript: str, *, title: str | None = None) -> SummaryResult:
+        payload = {
+            "model": self.model,
+            "stream": False,
+            "format": "json",
+            # qwen3-family models "think" by default: a long reasoning
+            # chain before the JSON answer, brutally slow on CPU and it
+            # does not improve the summary. Ask for the answer directly
+            # (ignored by non-thinking models).
+            "think": False,
+            "options": {"num_ctx": 8192},
+            "messages": [
+                {"role": "system", "content": SYSTEM_PROMPT},
+                {"role": "user", "content": build_user_message(transcript, title)},
+            ],
+        }
+        try:
+            async with self._client() as client:
+                resp = await client.post(f"{self.base_url}/api/chat", json=payload)
+        except (httpx.TimeoutException, httpx.TransportError) as e:
+            raise ProviderTransientError(f"Ollama unreachable: {type(e).__name__}") from e
+        if resp.status_code == 404:
+            # Missing model and missing route both 404 here; both are
+            # configuration, not weather.
+            raise ProviderConfigError("Ollama has no such model or route (HTTP 404).")
+        if resp.status_code != 200:
+            raise ProviderTransientError(f"Summarization failed (HTTP {resp.status_code}).")
+        try:
+            content = resp.json()["message"]["content"]
+        except (ValueError, KeyError, TypeError) as e:
+            raise ProviderTransientError("Ollama sent an unreadable reply.") from e
+        import json as _json
+
+        try:
+            data = _json.loads(content)
+        except ValueError as e:
+            raise ProviderTransientError("Ollama reply was not JSON.") from e
+        return parse_summary(data, self.model)