fix(contacts): surface ORG/TITLE/NOTE/URL/CATEGORIES/PHOTO on read (refs #716)

PR #719 fixed the contact-create path so all documented fields persist to the
vCard, but the read path (list/search via MCP) still returned ``organization:
null`` / ``note: null`` / ``title: null`` because pythonvCard4 has no typed
parser for ORG/TITLE — they land in ``Contact.custom`` — and the server-side
mapper never read ``note``/``urls``/``categories``/``photo`` even when present.

Reads now surface what the write side persisted:

- ``client/contacts.py``: new ``_first_custom`` helper pulls raw values from
  ``Contact.custom`` for ORG/TITLE/unencoded PHOTO. ``list_contacts``
  extends its per-contact dict with org/title/note/url/categories/photo.
- ``server/contacts.py``: ``_raw_contact_to_model`` maps the new keys onto
  ``Contact.organization`` / ``.title`` / ``.note`` / ``.urls`` / ``.categories``
  / ``.photo``. URL accepts both list and plain-string shapes; categories
  accepts comma-separated strings for forward-compat.

Coverage:

- Unit: ``TestFirstCustom`` (five cases incl. bare-string library shape) and
  three new ``_raw_contact_to_model`` cases covering the full field set,
  plain-string URL, and comma-string categories.
- Integration: ``test_mcp_contacts_workflow`` now decodes the
  ``nc_contacts_search_contacts`` response and asserts
  ``organization`` / ``note`` round-trip — direct regression coverage for
  elvisdragonmao's report on issue #716.

Verified end-to-end against the local single-user docker stack: creating a
contact with ``{organization, title, note, url, categories}`` and reading it
back via ``nc_contacts_search_contacts`` returns every field populated.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tests/server/test_contacts_mcp.py

@@ -1,5 +1,6 @@
 """Integration tests for Contacts MCP tools."""
 
+import json
 import logging
 import uuid
 
@@ -12,6 +13,11 @@ logger = logging.getLogger(__name__)
 pytestmark = pytest.mark.integration
 
 
+def _extract_payload(tool_result) -> dict:
+    """Return the JSON-decoded text content of an MCP tool result."""
+    return json.loads(tool_result.content[0].text)
+
+
 async def test_mcp_contacts_workflow(
     nc_mcp_client: ClientSession, nc_client: NextcloudClient
 ):
@@ -62,6 +68,23 @@ async def test_mcp_contacts_workflow(
         assert "ORG:MCP Test Corp" in raw_vcard
         assert f"NOTE:Created by test {unique_suffix}" in raw_vcard
 
+        # 4a. Read-side round-trip — issue #716 follow-up. The write side has
+        # been correct since PR #719, but the MCP list/search tools returned
+        # ``organization: null`` / ``note: null`` because pythonvCard4 stashes
+        # ORG/TITLE in ``custom`` and the server's _raw_contact_to_model never
+        # surfaced ``note`` / ``urls`` either.
+        search_result = await nc_mcp_client.call_tool(
+            "nc_contacts_search_contacts",
+            {"query": unique_suffix, "addressbook": addressbook_name},
+        )
+        assert search_result.isError is False
+        search_payload = _extract_payload(search_result)
+        assert search_payload["total_count"] == 1
+        searched = search_payload["contacts"][0]
+        assert searched["uid"] == contact_uid
+        assert searched["organization"] == "MCP Test Corp"
+        assert searched["note"] == f"Created by test {unique_suffix}"
+
         # 4b. Update with a URL — regression guard for PR #719 review:
         # _merge_vcard_properties previously had no URL handler, silently dropping it.
         update_result = await nc_mcp_client.call_tool(