feat(US1): text triple extraction — POST /api/v1/text/extract

- app/models/text_models.py: TripleItem, SourceOffset, TextExtract{Request,Response}
- app/services/text_service.py: TXT/PDF/DOCX parsing + LLM call + JSON parse
- app/routers/text.py: POST /text/extract handler with Depends injection
- tests/test_text_service.py: 6 unit tests (formats, errors)
- tests/test_text_router.py: 4 router tests (200, 400, 502×2)
- 10/10 tests passing

app/models/text_models.py

@@ -0,0 +1,25 @@
+from pydantic import BaseModel
+
+
+class SourceOffset(BaseModel):
+    start: int
+    end: int
+
+
+class TripleItem(BaseModel):
+    subject: str
+    predicate: str
+    object: str
+    source_snippet: str
+    source_offset: SourceOffset
+
+
+class TextExtractRequest(BaseModel):
+    file_path: str
+    file_name: str
+    model: str | None = None
+    prompt_template: str | None = None
+
+
+class TextExtractResponse(BaseModel):
+    items: list[TripleItem]