presence: Use the narrow user cache.

These two endpoints make up ~85% of requests to Zulip servers; since
presence is also a performance-critical endpoint, having them share
the same cache increases how hot it stays in memcached, in addition to
making the presence endpoint faster.

This comes at the very slightly cost of one extra field.  Checks
for the `is_bot` column are switched to the equivalent `bot_type`
check, since the columns are slightly duplicative, and we can get away
with only checking bot_type.

zerver/tests/test_presence.py

@@ -679,6 +679,31 @@ class UserPresenceTests(ZulipTestCase):
         result_dict = self.assert_json_success(result)
         self.assertEqual(set(result_dict["presences"].keys()), {othello.email})
 
+    def test_query_counts(self) -> None:
+        self.login("hamlet")
+        with self.assert_database_query_count(6):
+            # 1. session
+            # 2. narrow user cache
+            # 3. client
+            # 4. lock the userpresence row
+            # 5. update the userpresence row
+            # 6. select other userpresence data
+            self.assert_json_success(
+                self.client_post("/json/users/me/presence", {"status": "active"})
+            )
+
+        with self.assert_database_query_count(3, keep_cache_warm=True):
+            # With a warm cache, we skip the first three queries
+            self.assert_json_success(
+                self.client_post("/json/users/me/presence", {"status": "active"})
+            )
+
+        with self.assert_database_query_count(3, keep_cache_warm=True):
+            # It's the same story if we're becoming idle, as well
+            self.assert_json_success(
+                self.client_post("/json/users/me/presence", {"status": "idle"})
+            )
+
 
 class SingleUserPresenceTests(ZulipTestCase):
     def test_email_access(self) -> None: