https://github.com/camsaul/toucan2/pull/126

Editing the data model loads a _lot_ of data. It hits the following form
from the api:

```clojure
(hydrate [:tables [:fields [:target :has_field_values]
                   :has_field_values] :segments :metrics])
```

On my test instances, with 1500 tables each with 14 fields: 15mb and 359
tables each with 14 fields: 3.69mb. Not great.

If you want to recreate, create an empty db, sync it, then

```clojure
  (time
   (let [conn (sql-jdbc.conn/db->pooled-connection-spec 3)]
     (dotimes [n 359]
       (jdbc/execute! conn [(format "create table if not exists foo_%d (id int, col_1 text, col_2 text, col_3 text, col_4 text, col_5 text, col_6 text, col_7 text, col_8 text, col_9 text, col_10 text, col_11 text, col_12 text, col_13 text, col_14 text)" n)]))))

  (sync/sync-database! (db/select-one 'Database :id 3) {:scan :schema})

  (binding [api/*current-user-permissions-set* (delay #{"/"})]
    (count (#'metabase.api.database/db-metadata 3 true true)))
```

And the source of the bug was a classic recur with a lazy concat.

```diff
diff --git a/src/toucan2/tools/hydrate.clj b/src/toucan2/tools/hydrate.clj
index e1d0459..10c9c49 100644
--- a/src/toucan2/tools/hydrate.clj
+++ b/src/toucan2/tools/hydrate.clj
@@ -456,7 +456,7 @@
     (if (empty? hydrated-instances)
       (concat acc (map :instance annotated-instances))
       (let [[not-hydrated [_needed-hydration & more]] (split-with (complement :needs-hydration?) annotated-instances)]
-        (recur (concat acc (map :instance not-hydrated) [(first hydrated-instances)])
+        (recur (vec (concat acc (map :instance not-hydrated) [(first hydrated-instances)]))
                more
                (rest hydrated-instances))))))

```