Add gpt-5.4 to getcurrent snapshot to avoid cross-test model cache contamination

Both tests in rpc_session_state.e2e.test.ts share one long-lived CLI subprocess (one client created at describe scope). The runtime caches the /models response from whichever test calls it first. When 'getcurrent' ran first with a single-model snapshot, that response was cached; then 'switchto' would accept gpt-5.4 synchronously but the background tool-revalidation triggered by the model_change event saw only claude-sonnet-4.5 in the cached list, rejected gpt-5.4, and silently reverted _selectedModel. Listing both models in the getcurrent snapshot keeps the cached list compatible with switchto.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: SteveSandersonMS

test/snapshots/rpc_session_state/should_call_session_rpc_model_getcurrent.yaml

@@ -1,3 +1,4 @@
 models:
   - claude-sonnet-4.5
+  - gpt-5.4
 conversations: []