Stop LLM adapters following redirects and cap Retry-After

MQ-01: the Anthropic adapter used a default http.Client, so a 3xx from the
configured endpoint was followed and the x-api-key header resent to the
redirect target; net/http strips only its own sensitive headers. Both
adapters now use a client whose CheckRedirect returns ErrUseLastResponse,
so a 3xx comes back as an llm.ProviderError and nothing is requested again.

The OpenAI-compatible adapter gets the same change: it forwards configured
extra headers (which may carry a credential and which net/http does not
treat as sensitive) across hosts, and net/http keeps Authorization on a
same-host https-to-http redirect.

MQ-02: a delay-seconds Retry-After too large for time.Duration overflowed
into a negative wait. retryAfter now caps the value at the most whole
seconds a time.Duration holds (also for values beyond int64) and the
response still maps to ErrRateLimited.

Tests: TestRedirectIsNotFollowed (302 to a second server, which must get
no request) and TestRateLimitedHugeRetryAfterIsCapped in both packages,
plus huge-number cases in TestRetryAfter.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

internal/ai/llm/openai/openai_test.go

@@ -779,6 +779,9 @@ func TestRetryAfter(t *testing.T) {
 		"soon":                          0,
 		"Sat, 03 Oct 2026 12:02:00 GMT": 2 * time.Minute,
 		"Sat, 03 Oct 2026 11:59:00 GMT": 0,
+		"9223372036854775807":           time.Duration(maxRetryAfterSeconds) * time.Second,
+		"99999999999999999999":          time.Duration(maxRetryAfterSeconds) * time.Second,
+		"-99999999999999999999":         0,
 	}
 	for in, want := range cases {
 		if got := retryAfter(in, now); got != want {
@@ -1142,3 +1145,60 @@ func TestRedactedErrorsFormatWithoutKey(t *testing.T) {
 		}
 	}
 }
+
+func TestRateLimitedHugeRetryAfterIsCapped(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Retry-After", "9223372036854775807")
+		w.WriteHeader(http.StatusTooManyRequests)
+	}))
+	t.Cleanup(srv.Close)
+	p, err := New(Config{BaseURL: srv.URL + "/v1", APIKey: testKey, Model: "test-model", Headers: map[string]string{"X-Title": testKey}, StructuredOutput: PromptOnly})
+	if err != nil {
+		t.Fatal(err)
+	}
+	err = p.Verify(context.Background())
+	var rl *llm.RateLimitedError
+	if !errors.Is(err, llm.ErrRateLimited) || !errors.As(err, &rl) || rl.RetryAfter != time.Duration(maxRetryAfterSeconds)*time.Second {
+		t.Fatalf("err = %v, want a RateLimitedError with RetryAfter capped to %v", err, time.Duration(maxRetryAfterSeconds)*time.Second)
+	}
+}
+
+func TestRedirectIsNotFollowed(t *testing.T) {
+	var mu sync.Mutex
+	var secondCalls int
+	var secondHeader http.Header
+	second := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
+		mu.Lock()
+		secondCalls++
+		secondHeader = req.Header.Clone()
+		mu.Unlock()
+		w.WriteHeader(http.StatusOK)
+	}))
+	t.Cleanup(second.Close)
+	first := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
+		http.Redirect(w, req, second.URL+req.URL.Path, http.StatusFound)
+	}))
+	t.Cleanup(first.Close)
+	p, err := New(Config{BaseURL: first.URL + "/v1", APIKey: testKey, Model: "test-model", Headers: map[string]string{"X-Title": testKey}, StructuredOutput: PromptOnly})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = p.Verify(context.Background())
+	var pe *llm.ProviderError
+	if !errors.Is(err, llm.ErrProvider) || !errors.As(err, &pe) || pe.Status() != http.StatusFound {
+		t.Fatalf("err = %v, want a ProviderError for HTTP 302", err)
+	}
+	mu.Lock()
+	defer mu.Unlock()
+	if secondCalls != 0 {
+		for name, values := range secondHeader {
+			for _, v := range values {
+				if strings.Contains(v, testKey) {
+					t.Errorf("redirect target saw the key in %s", name)
+				}
+			}
+		}
+		t.Fatalf("redirect target got %d requests, want 0", secondCalls)
+	}
+}