feat(rag): PDF/docx ingest + chat citations

- Upload endpoint (base64 JSON, no multipart dep): extracts text from
  pdf/docx/txt/md via pypdf + python-docx, then runs the existing
  chunk/embed pipeline. Documents page uploads files straight through.
- Citations: the chat stream emits an SSE `sources` event listing the
  documents that fed the answer; the UI shows them as chips under the reply.
- Deps: pypdf, python-docx (both pure-Python, Windows-safe).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

tests/test_documents.py

@@ -55,3 +55,19 @@ def test_add_list_search_delete_roundtrip():
 def test_search_empty_query_returns_nothing():
     s = _store()
     assert asyncio.run(s.search_documents("", _fake_embed)) == []
+
+
+def test_extract_text_by_type():
+    from synapse.main import _extract_text
+    # plain text / markdown -> UTF-8 decode
+    assert _extract_text("notes.md", b"# Title\n\nbody") == "# Title\n\nbody"
+    assert _extract_text("x.txt", "café".encode("utf-8")) == "café"
+    # a real (tiny) PDF built with pypdf -> text extracted back out
+    from pypdf import PdfWriter, PdfReader
+    import io
+    w = PdfWriter()
+    w.add_blank_page(width=200, height=200)
+    buf = io.BytesIO(); w.write(buf)
+    out = _extract_text("blank.pdf", buf.getvalue())
+    assert isinstance(out, str)  # blank page -> "" or whitespace, never raises
+    assert PdfReader(io.BytesIO(buf.getvalue())).pages  # sanity: it was a valid PDF