update_node with ws URL hangs forever when re-subscribing after pub/sub disable

`update_node` using a `url` (wss://...) instead of `node_name` hangs indefinitely when re-enabling subscriptions that were previously disabled.

## Reproduction

1. Set `subscribe: false, publish: false` on a node via `update_node { url: "wss://..." }`.
2. Insert records locally.
3. Set `subscribe: true, publish: true` via `update_node { url: "wss://..." }` again.

**Result:** the request hangs until the server-side timeout (tested up to 35 minutes).

**Using `node_name` instead of `url` does not hang** — the bug is specific to URL-based `update_node`.

## Notes

- Observed consistently across multiple CI runs.
- The `subscriptions` array format in the reproduction uses the legacy replication API (pre-5.x). Priority may change once the legacy path is fully removed.

---

🤖 Filed by Claude on behalf of Kris.

Start by running the two-step reproduction against update_node with the wss:// URL, then compare its re-subscription path with the node_name path. Trace the URL-based update_node handling and the legacy subscriptions path to find where the request stops progressing. Done means re-enabling subscribe and publish returns without hanging and the regression is covered by a repeatable test.

Written by the indexing model from the issue text.