Person page: server-side search; stop loading the whole tree

The person page fetched the entire tree on every open — all persons (to build a
name map + power the relative pickers) and all events (to find partnership
events). On a 2k-person tree that's a ~230KB person list + ~600KB event list per
view. Now it loads only what the page shows:

Frontend:
- The relationship & spouse pickers use the backend's fuzzy pg_trgm search
  (debounced, typo-tolerant) instead of substring-filtering a preloaded array —
  better search, and no need to preload every person. PersonCombobox gained an
  `onSearch` server mode (client `people` mode still works).
- The page drops the all-persons and all-events fetches; it resolves just this
  person's relatives' names via GET /persons?ids=..., and reads partnership
  events from the per-person events endpoint.

Backend:
- GET /trees/{id}/persons?ids=a,b,c — batch by id (privacy-filtered, names
  batched), for relative-name display.
- list_events_for_person (member path) now also returns the person's partnership
  events, so the page needn't scan every event in the tree.

Adversarial review (frontend logic + backend/privacy) found no issues. Suite 105
passing.

Signed-off-by: Justin Paul <justin@jpaul.me>

frontend/openapi.json

@@ -851,6 +851,22 @@
               ],
               "title": "Q"
             }
+          },
+          {
+            "name": "ids",
+            "in": "query",
+            "required": false,
+            "schema": {
+              "anyOf": [
+                {
+                  "type": "string"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "title": "Ids"
+            }
           }
         ],
         "responses": {