fix(model-error-classifier): mark OpenAI server_error patterns as retryable (fixes #3799)

OpenAI streaming responses can surface mid-stream errors with type
'server_error' or with the prose message 'An error occurred while processing
your request'. Neither matched any entry in RETRYABLE_MESSAGE_PATTERNS, so
shouldRetryError returned false and the runtime-fallback / fallback-retry
code paths skipped retry. The result was that GPT-5.5 subagent (and main)
turns silently stalled until the stale timeout fired.

The maintainer's diagnosis on issue #3799 explicitly recommends adding
these two patterns to model-error-classifier.ts; this commit does exactly
that and adds two regression tests covering the JSON envelope and the
prose form.

src/shared/model-error-classifier.test.ts

@@ -434,6 +434,34 @@ describe("model-error-classifier", () => {
     //#then
     expect(result).toBe(true)
   })
+
+  test("treats OpenAI streaming server_error envelopes as retryable (issue #3799)", () => {
+    //#given: OpenAI surfaces its mid-stream error with type 'server_error'
+    const error = {
+      name: undefined,
+      message: "{\"error\":{\"type\":\"server_error\",\"message\":\"server_error\"}}",
+    }
+
+    //#when
+    const result = shouldRetryError(error)
+
+    //#then
+    expect(result).toBe(true)
+  })
+
+  test("treats the OpenAI prose 'An error occurred while processing' message as retryable (issue #3799)", () => {
+    //#given: the human-readable prose surfaced when OpenAI's stream fails
+    const error = {
+      name: undefined,
+      message: "An error occurred while processing your request. Please try again later.",
+    }
+
+    //#when
+    const result = shouldRetryError(error)
+
+    //#then
+    expect(result).toBe(true)
+  })
 })
 
 export {}