fix(mistral): honor dataset language in figure prompts (#18021)

### Summary

Refs #17885.

Mistral figure enrichment now receives the dataset language through the
production parsing path. `by_mistral_ocr` forwards `lang` to
`MistralParser.parse_pdf`; the parser stores the normalized language and
passes it to the figure-description prompt. Empty or missing values
still fall back to English.

test/unit_test/deepdoc/parser/test_mistral_parser.py

@@ -539,8 +539,11 @@ def test_parse_pdf_consumes_vision_model_kwarg(monkeypatch, tmp_path):
     _patch_render(m, p, 2)
     pdf = tmp_path / "x.pdf"
     pdf.write_bytes(b"%PDF-1.4 minimal")
-    p.parse_pdf(str(pdf), vision_model="VM")
+    p.parse_pdf(str(pdf), vision_model="VM", lang="Japanese")
     assert p.vision_model == "VM"  # popped from kwargs into self, not forwarded to _call_ocr
+    assert p.language == "Japanese"
+    p.parse_pdf(str(pdf), vision_model="VM", lang="")
+    assert p.language == "English"
 
 
 def test_describe_image_passes_pil_image_not_bytes(monkeypatch):
@@ -562,10 +565,11 @@ def test_describe_image_passes_pil_image_not_bytes(monkeypatch):
     pic = ModuleType("rag.app.picture")
     pic.vision_llm_chunk = lambda binary, vision_model, prompt=None, callback=None: (captured.update(kind=type(binary).__name__), "a white square")[1]
     gen = ModuleType("rag.prompts.generator")
-    gen.vision_llm_figure_describe_prompt = lambda: "describe"
+    gen.vision_llm_figure_describe_prompt = lambda language: (captured.update(language=language), "describe")[1]
     monkeypatch.setitem(_sys.modules, "rag.app.picture", pic)
     monkeypatch.setitem(_sys.modules, "rag.prompts.generator", gen)
 
     out = p._describe_image("@@1\t0\t0\t40\t40##")
     assert out == "a white square"
     assert captured["kind"] == "Image"  # PIL Image, not 'bytes'
+    assert captured["language"] == "English"

test/unit_test/rag/app/test_vision_language_callers.py

@@ -98,3 +98,30 @@ def test_markdown_chunk_forwards_language_to_model_and_figure_parser(monkeypatch
     assert parser_factory.call_args.kwargs["vision_model"] is vision_model
     assert parser_factory.call_args.kwargs["lang"] == "Japanese"
     parser_instance.assert_called_once()
+
+
+@pytest.mark.p1
+def test_mistral_ocr_forwards_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": "mistral-ocr"}))
+    monkeypatch.setattr(naive, "LLMBundle", Mock(return_value=ocr_model))
+
+    sections, tables, returned_parser = naive.by_mistral_ocr(
+        "document.pdf",
+        binary=b"pdf",
+        from_page=2,
+        to_page=5,
+        lang="Japanese",
+        callback=lambda *_args, **_kwargs: None,
+        parse_method="raw",
+        mistral_ocr_llm_name="mistral-ocr",
+        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"