revert #386 due to unsafe shell command parsing (#478)

Reverts https://github.com/openai/codex/pull/386 because:

* The parsing logic for shell commands was unsafe (`split(/\s+/)`
instead of something like `shell-quote`)
* We have a different plan for supporting auto-approved commands.

codex-cli/tests/approvals.test.ts

@@ -1,13 +1,7 @@
 import type { SafetyAssessment } from "../src/approvals";
 
 import { canAutoApprove } from "../src/approvals";
-import { describe, test, expect, vi } from "vitest";
-
-vi.mock("../src/utils/config", () => ({
-  loadConfig: () => ({
-    safeCommands: ["npm test", "sl"],
-  }),
-}));
+import { describe, test, expect } from "vitest";
 
 describe("canAutoApprove()", () => {
   const env = {
@@ -95,27 +89,4 @@ describe("canAutoApprove()", () => {
 
     expect(check(["cargo", "build"])).toEqual({ type: "ask-user" });
   });
-
-  test("commands in safeCommands config should be safe", async () => {
-    expect(check(["npm", "test"])).toEqual({
-      type: "auto-approve",
-      reason: "User-defined safe command",
-      group: "User config",
-      runInSandbox: false,
-    });
-
-    expect(check(["sl"])).toEqual({
-      type: "auto-approve",
-      reason: "User-defined safe command",
-      group: "User config",
-      runInSandbox: false,
-    });
-
-    expect(check(["npm", "test", "--watch"])).toEqual({
-      type: "auto-approve",
-      reason: "User-defined safe command",
-      group: "User config",
-      runInSandbox: false,
-    });
-  });
 });