feat(ocr): observe ocr_processing_seconds around engine.to_thread calls

Wraps every asyncio.to_thread(engine.extract_*) call with time.monotonic()
deltas in /ocr (per document) and in both /ocr/stream generators (per page).
Streaming buckets are the useful operational signal; the non-streaming
observation is a bonus.

Refs #652 (AC5)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ocr-service/test_metrics.py

@@ -267,3 +267,43 @@ async def test_ocr_words_and_illegible_words_total_sum_across_blocks(fresh_metri
 
     assert fresh_metrics.ocr_words_total._value.get() == 5.0
     assert fresh_metrics.ocr_illegible_words_total._value.get() == 2.0
+
+
+def _histogram_count_sum(histogram, **labels) -> tuple[float, float]:
+    """Read the per-label-set _count and _sum from a prometheus_client Histogram."""
+    child = histogram.labels(**labels)
+    return child._sum.get(), sum(b.get() for b in child._buckets)
+
+
+@pytest.mark.asyncio
+async def test_ocr_processing_seconds_histogram_observed_per_page_in_stream(fresh_metrics):
+    """The streaming generator observes ocr_processing_seconds once per page."""
+    mock_images = [Image.new("RGB", (100, 100)) for _ in range(2)]
+    mock_blocks = [{"pageNumber": 1, "x": 0.0, "y": 0.0, "width": 1.0, "height": 1.0,
+                    "polygon": None, "text": "ok", "words": []}]
+
+    with patch("main.kraken_engine.load_models"), \
+         patch("main.load_spell_checker"), \
+         patch("main._download_and_convert_pdf", new_callable=AsyncMock, return_value=mock_images), \
+         patch("main.preprocess_page", side_effect=lambda img: img), \
+         patch("main.surya_engine.extract_page_blocks", return_value=mock_blocks):
+        async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
+            import main as main_module
+            main_module._models_ready = True
+            try:
+                async with client.stream("POST", "/ocr/stream", json={
+                    "pdfUrl": "http://minio/doc.pdf",
+                    "scriptType": "TYPEWRITER",
+                    "language": "de",
+                }) as response:
+                    assert response.status_code == 200
+                    async for _ in response.aiter_lines():
+                        pass
+            finally:
+                main_module._models_ready = False
+
+    sum_seconds, count = _histogram_count_sum(
+        fresh_metrics.ocr_processing_seconds, engine="surya"
+    )
+    assert count == 2.0
+    assert sum_seconds >= 0.0