fix(mineru): honor dataset language in figure prompts (#18188)

### Summary

Closes #17885.

MinerU figure enrichment now passes the resolved dataset language to
`vision_llm_figure_describe_prompt`. Missing and empty language values
use `English`, matching the other figure-description paths.

This change is limited to MinerU. PR #18021 already fixed the Mistral
path.

test/unit_test/rag/app/test_vision_language_callers.py

@@ -125,3 +125,30 @@ def test_mistral_ocr_forwards_language_to_parser(monkeypatch):
     assert tables == []
     assert returned_parser is parser
     assert parser.parse_pdf.call_args.kwargs["lang"] == "Japanese"
+
+
+@pytest.mark.p1
+def test_mineru_forwards_dataset_language_to_parser(monkeypatch):
+    parser = Mock()
+    parser.parse_pdf.return_value = (["section"], [])
+    ocr_model = Mock(mdl=parser)
+    monkeypatch.setattr(naive, "resolve_model_config", Mock(return_value={"llm_name": "mineru"}))
+    monkeypatch.setattr(naive, "LLMBundle", Mock(return_value=ocr_model))
+
+    sections, tables, returned_parser = naive.by_mineru(
+        "document.pdf",
+        binary=b"pdf",
+        from_page=2,
+        to_page=5,
+        lang="Japanese",
+        callback=lambda *_args, **_kwargs: None,
+        parse_method="raw",
+        mineru_llm_name="mineru",
+        tenant_id="tenant-id",
+        vision_model=object(),
+    )
+
+    assert sections == ["section"]
+    assert tables == []
+    assert returned_parser is parser
+    assert parser.parse_pdf.call_args.kwargs["lang"] == "Japanese"