fix(agents): deny apply_patch for GPT models to prevent verification hangs (#2935)

GPT models (5.3-codex, 5.4, etc.) frequently hang when using apply_patch
due to verification loops. This adds:

1. Tool restriction: apply_patch is denied for GPT variants of
   Hephaestus, Sisyphus-Junior, and Sisyphus agents
2. Prompt guidance: GPT-specific prompts now explicitly instruct using
   edit/write tools instead of apply_patch
3. Removed the 'Always use apply_patch' instruction from
   sisyphus-junior/gpt-5-4.ts that contradicted the fix

The deny is model-conditional — Claude variants retain apply_patch
access since it works reliably there.

src/agents/delegation-trust-prompt.test.ts

@@ -114,6 +114,8 @@ describe("delegation trust prompt rules", () => {
     expect(prompt).toContain("do only non-overlapping work simultaneously")
     expect(prompt).toContain("Continue only with non-overlapping work")
     expect(prompt).toContain("DO NOT perform the same search yourself")
+    expect(prompt).toContain("Do not use `apply_patch`")
+    expect(prompt).toContain("`edit` and `write`")
   })
 
   test("Sisyphus-Junior GPT-5.4 prompt forbids duplicate delegated exploration", () => {