fix: corrige bugs críticos de índice de páginas e cache de OCR

- C-1: corrige índice de páginas OCR'd (retorna page_number em vez de position)
- C-2: adiciona tratamento para falhas de OCR no cache (não distingue erro de não processado)
- Integra paralelização do OCR via subprocessos (parallel.py)
- Indexador usa extração paralela quando disponível
- Melhora relatório de progresso com contagem de páginas processadas

backend/app/indexer.py

@@ -208,6 +208,51 @@ class Indexer:
 
             return self.status()
 
+    def _extract_pages(self, pdf: Path, file_hash: str) -> list:
+        """Extrai as páginas do PDF, usando OCR paralelo quando possível."""
+        extract_parallel = getattr(self.extractor, "extract_pages_parallel", None)
+        if extract_parallel is None:
+            return list(self.extractor.extract_document(pdf, file_hash=file_hash))
+
+        page_count = _pdf_page_count(pdf)
+        pages: list = []
+        batch_size = max(self.settings.ocr_workers, 1) * 4
+
+        for batch in self._page_batches(pdf, batch_size):
+            raw_pages = [
+                (page_number, text)
+                for page_number, text in batch
+            ]
+            extracted = extract_parallel(
+                raw_pages,
+                file_hash=file_hash,
+                path=pdf,
+                workers=self.settings.ocr_workers,
+            )
+            pages.extend(extracted)
+            self.report_progress(pdf.name, len(pages), page_count)
+
+        return sorted(pages, key=lambda page: page.page)
+
+    def _page_batches(self, pdf: Path, batch_size: int):
+        """Gera lotes de (número_da_página, texto_da_camada)."""
+        import pymupdf
+
+        with pymupdf.open(pdf) as document:
+            batch: list[tuple[int, str]] = []
+            for index in range(len(document)):
+                page = document[index]
+                try:
+                    text = page.get_text("text") or ""
+                except Exception:  # pragma: no cover - PDF malformado
+                    text = ""
+                batch.append((index + 1, text))
+                if len(batch) >= batch_size:
+                    yield batch
+                    batch = []
+            if batch:
+                yield batch
+
     def _index_file(self, pdf: Path, force: bool) -> None:
         file_hash = file_sha256(pdf)
         entry = self._manifest.get(pdf.name)
@@ -217,15 +262,7 @@ class Indexer:
             return
 
         logger.info("Indexando %s", pdf.name)
-        pages = []
-        page_count = _pdf_page_count(pdf)
-        for position, page in enumerate(
-            self.extractor.extract_document(pdf, file_hash=file_hash), start=1
-        ):
-            pages.append(page)
-            if position % 10 == 0 or position == page_count:
-                self.report_progress(pdf.name, position, page_count)
-
+        pages = self._extract_pages(pdf, file_hash)
         failed_pages = [page.page for page in pages if page.failed]
 
         book_starts = self._book_starts(pages)