test(vcl): start VCLBridge in vcl_executor_test setup_all (fix seed-order flake)

Unrelated to the NIF change in this PR, but surfaced by this PR's CI seed
(590007). VCLExecutorTest's execute_string/2 tests route through the
VeriSim.Query.VCLBridge parser singleton via GenServer.call by name, but
this file — alone among the VCL test files — never started that GenServer
in its setup. It therefore passed only when a bridge-starting sibling test
(which use H.ensure_bridge_started/0 or VCLBridge.start_link/1) happened to
run earlier; an unlucky seed ordered VCLExecutorTest first and the three
parse-error-propagation tests exited with 'no process'.

Add a setup_all that calls the existing VCLTestHelpers.ensure_bridge_started/0
(idempotent: returns the already-started pid), matching every sibling VCL
test file, so the suite is order-independent. Test-only; does not touch the
in-flight VCL rename of lib/ modules.

https://claude.ai/code/session_01E8BpV19yxhf67UrCrvkfTr

Author: claude

elixir-orchestration/test/verisim/query/vcl_executor_test.exs

@@ -24,6 +24,17 @@ defmodule VeriSim.Query.VCLExecutorTest do
   use ExUnit.Case, async: false
 
   alias VeriSim.Query.VCLExecutor
+  alias VeriSim.Test.VCLTestHelpers, as: H
+
+  setup_all do
+    # `execute_string/2` routes through the VCLBridge parser singleton. Every
+    # other VCL test file starts that GenServer in its own setup; this file did
+    # not, so its execute_string tests passed only when a bridge-starting file
+    # happened to run earlier (a seed-order dependency). Start it explicitly so
+    # the suite is order-independent.
+    H.ensure_bridge_started()
+    :ok
+  end
 
   describe "execute/2 with :explain option" do
     test "returns a structured plan with the documented keys" do