feat(ocr): time only engine work in guided stream histogram

Previously the guided generator's page_started timer wrapped the entire
region loop including the synchronous correct_text() call, inflating
ocr_processing_seconds with spell-check latency. Sum the per-region
engine.extract_region_text durations instead so the histogram matches
the unguided stream's "engine only" semantic.

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

ocr-service/main.py

@@ -217,13 +217,15 @@ async def run_ocr_stream(request: OcrRequest):
                     image = await asyncio.to_thread(preprocess_page, image)
                     blocks = []
                     sender_path = request.senderModelPath if use_kraken else None
-                    page_started = time.monotonic()
+                    engine_seconds = 0.0
                     for region in page_regions:
+                        region_started = time.monotonic()
                         text = await asyncio.to_thread(
                             engine.extract_region_text, image,
                             region.x, region.y, region.width, region.height,
                             sender_path,
                         )
+                        engine_seconds += time.monotonic() - region_started
                         if script_type in _SPELL_CHECK_SCRIPT_TYPES:
                             text = correct_text(text)
                         blocks.append({
@@ -238,7 +240,7 @@ async def run_ocr_stream(request: OcrRequest):
                         })
 
                     metrics.ocr_processing_seconds.labels(engine=engine_name).observe(
-                        time.monotonic() - page_started
+                        engine_seconds
                     )
                     total_blocks += len(blocks)
                     metrics.ocr_pages_total.labels(engine=engine_name).inc()