feat(ocr): PDF 文本层可用时跳过 worker 调用并补装 poppler-data

- OcrService 提取 PDF 文本层后若有效字符达到阈值，直接构建文档并写入结果缓存，不再触发 OCR worker，仅无文本层时才解析 python_bin/worker_path 调用 worker
- _build_text_layer_document 复用 AggregatedOcrDocument 聚合文本层片段，_has_usable_pdf_text_layer 基于 meaningful_char_count 判定
- docker-compose 与 paddleocr bootstrap 脚本补装 poppler-data，保证 PDF 文本层抽取的中文编码正确
- 新增文本层直取与运行时依赖两项 ocr_service 单测

server/src/app/services/ocr.py

@@ -77,8 +77,6 @@ class OcrService:
         documents: list[OcrRecognizeDocumentRead] = []
         prepared_inputs: list[PreparedOcrInput] = []
         cleanup_paths: list[Path] = []
-        python_bin = self._resolve_python_bin()
-        worker_path = self._resolve_worker_path()
         worker_payload: dict = {}
         cache_keys_by_source: dict[str, str] = {}
 
@@ -144,6 +142,16 @@ class OcrService:
                             cleanup_paths=cleanup_paths,
                             text_layer=text_layer,
                         )
+                        if self._has_usable_pdf_text_layer(text_layer):
+                            document = self._build_text_layer_document(
+                                filename=normalized_name,
+                                media_type=resolved_media_type,
+                                text_layer=text_layer,
+                                pdf_inputs=pdf_inputs,
+                            )
+                            documents.append(document)
+                            self._write_cached_document(cache_key, document)
+                            continue
                         prepared_inputs.extend(pdf_inputs)
                         for item in pdf_inputs:
                             cache_keys_by_source.setdefault(item.source_key, cache_key)
@@ -175,6 +183,8 @@ class OcrService:
                 cache_keys_by_source[source_key] = cache_key
 
             if prepared_inputs:
+                python_bin = self._resolve_python_bin()
+                worker_path = self._resolve_worker_path()
                 worker_payload = self._invoke_worker(
                     python_bin=python_bin,
                     worker_path=worker_path,
@@ -308,6 +318,23 @@ class OcrService:
             while len(cls._result_cache) > OCR_RESULT_CACHE_LIMIT:
                 cls._result_cache.popitem(last=False)
 
+    @classmethod
+    def _write_cached_document(cls, cache_key: str, document: OcrRecognizeDocumentRead) -> None:
+        if not cache_key:
+            return
+        with cls._cache_lock:
+            cls._result_cache[cache_key] = document.model_copy(
+                update={
+                    "receipt_id": "",
+                    "receipt_status": "",
+                    "receipt_preview_url": "",
+                    "receipt_source_url": "",
+                }
+            )
+            cls._result_cache.move_to_end(cache_key)
+            while len(cls._result_cache) > OCR_RESULT_CACHE_LIMIT:
+                cls._result_cache.popitem(last=False)
+
     @classmethod
     def _resolve_worker_semaphore(cls, limit: int) -> threading.Semaphore:
         normalized_limit = max(1, int(limit or 1))
@@ -568,6 +595,30 @@ class OcrService:
 
         return documents
 
+    def _build_text_layer_document(
+        self,
+        *,
+        filename: str,
+        media_type: str,
+        text_layer: str,
+        pdf_inputs: list[PreparedOcrInput],
+    ) -> OcrRecognizeDocumentRead:
+        first_input = pdf_inputs[0] if pdf_inputs else None
+        aggregated = AggregatedOcrDocument(
+            filename=filename,
+            media_type=media_type,
+            source_key=first_input.source_key if first_input is not None else uuid4().hex,
+            page_count=max(1, len(pdf_inputs)),
+            preview_kind=str(first_input.preview_kind if first_input is not None else ""),
+            preview_data_url=str(first_input.preview_data_url if first_input is not None else ""),
+        )
+        aggregated.text_layer_fragments.append(text_layer)
+        return self._finalize_document(aggregated)
+
+    @classmethod
+    def _has_usable_pdf_text_layer(cls, text_layer: str) -> bool:
+        return cls._meaningful_char_count(text_layer) >= 8
+
     @staticmethod
     def _collect_descriptor_text_layer(descriptors: list[PreparedOcrInput]) -> str:
         for descriptor in descriptors: