fix(grok): 405 纳入 failover 与临时踢号，修复 Grok 会话粘性锁死

Grok 分组里若有账号的自定义 base_url 只实现 /v1/chat/completions，对
/v1/responses 固定返回 405，会话 sticky 到该账号后就永久锁死：

1. shouldFailoverUpstreamError 不含 405，请求不进换号循环
2. handleGrokAccountUpstreamError 的 switch 不含 405，账号不进临时不可调度
3. 账号仍可调度，sticky 不清理，客户端自动重连继续绑同一个坏号

改动两处：
- shouldFailoverUpstreamError 增加 405，让请求可换号，sticky 随之更新
- handleGrokAccountUpstreamError 增加 405 → 30 分钟冷却，与 403 同级

Fixes #4668

backend/internal/service/openai_gateway_grok.go

@@ -1370,6 +1370,8 @@ func (s *OpenAIGatewayService) handleGrokAccountUpstreamError(ctx context.Contex
 		s.tempUnscheduleGrok(ctx, account, 30*time.Minute, "grok payment required")
 	case http.StatusForbidden:
 		s.tempUnscheduleGrok(ctx, account, 30*time.Minute, "grok access or entitlement denied")
+	case http.StatusMethodNotAllowed:
+		s.tempUnscheduleGrok(ctx, account, 30*time.Minute, "grok endpoint not supported (405)")
 	case http.StatusTooManyRequests:
 		// updateGrokUsageSnapshot installs rate-limit state for non-pool accounts.
 	default:

backend/internal/service/openai_gateway_grok_405_test.go

@@ -0,0 +1,29 @@
+package service
+
+import (
+	"net/http"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestShouldFailoverUpstreamError_405IsFailoverEligible(t *testing.T) {
+	svc := &OpenAIGatewayService{}
+
+	assert.True(t, svc.shouldFailoverUpstreamError(http.StatusMethodNotAllowed),
+		"405 should trigger failover so sticky sessions can escape to healthy accounts")
+}
+
+func TestShouldFailoverUpstreamError_ExistingCodesStillWork(t *testing.T) {
+	svc := &OpenAIGatewayService{}
+
+	failoverCodes := []int{401, 402, 403, 405, 429, 529, 500, 502, 503, 504}
+	for _, code := range failoverCodes {
+		assert.True(t, svc.shouldFailoverUpstreamError(code), "status %d should trigger failover", code)
+	}
+
+	nonFailoverCodes := []int{200, 201, 400, 404, 408, 422}
+	for _, code := range nonFailoverCodes {
+		assert.False(t, svc.shouldFailoverUpstreamError(code), "status %d should NOT trigger failover", code)
+	}
+}

backend/internal/service/openai_gateway_grok_test.go

@@ -2583,6 +2583,13 @@ func TestHandleGrokAccountUpstreamErrorTempUnschedulesNonRateLimitStates(t *test
 			wantMinCooldown: 2*time.Minute - time.Second,
 			wantMaxCooldown: 2*time.Minute + time.Second,
 		},
+		{
+			name:            "method not allowed",
+			status:          http.StatusMethodNotAllowed,
+			wantReason:      "grok endpoint not supported (405)",
+			wantMinCooldown: 30*time.Minute - time.Second,
+			wantMaxCooldown: 30*time.Minute + time.Second,
+		},
 	}
 
 	for _, tt := range tests {

backend/internal/service/openai_gateway_upstream_errors.go

@@ -211,7 +211,7 @@ func isOpenAIContextWindowError(upstreamMsg string, upstreamBody []byte) bool {
 
 func (s *OpenAIGatewayService) shouldFailoverUpstreamError(statusCode int) bool {
 	switch statusCode {
-	case 401, 402, 403, 429, 529:
+	case 401, 402, 403, 405, 429, 529:
 		return true
 	default:
 		return statusCode >= 500