fix(agent/title): disable thinking and raise max_tokens for Gemini

Gemini 2.5/3 default to high thinking via OpenAI-compat, consuming the
entire max_tokens budget and truncating titles to a single word (and
adding latency). Title generation is a trivial task that does not
benefit from reasoning, so set OptThinkingLevel="off" and bump
max_tokens from 50 to 256 for headroom across reasoning-capable
providers.

Also update mapGeminiReasoningEffort to forward "off" as "low" (the
minimum effort all Gemini models accept via OpenAI-compat), since not
forwarding causes the server to fall back to "high".

internal/agent/title_generate.go

@@ -26,8 +26,15 @@ func GenerateTitle(ctx context.Context, provider providers.Provider, model, user
 		},
 		Model: model,
 		Options: map[string]any{
-			providers.OptMaxTokens:   50,
-			providers.OptTemperature: 0.3,
+			// Larger budget: thinking-capable models (Gemini 2.5/3, GPT-5 reasoning)
+			// can consume output tokens on reasoning traces. 256 leaves room for a
+			// 15-word title even when the provider allocates some budget to thinking.
+			providers.OptMaxTokens:     256,
+			providers.OptTemperature:   0.3,
+			// Disable extended thinking for title generation — it's a trivial task
+			// that doesn't benefit from reasoning and defaults (esp. Gemini's "high")
+			// otherwise eat the entire max_tokens budget, truncating the title to 1 word.
+			providers.OptThinkingLevel: "off",
 		},
 	})
 	if err != nil {

internal/providers/openai_gemini_thinking_test.go

@@ -20,7 +20,7 @@ func TestBuildRequestBody_GeminiForwardsReasoningEffort(t *testing.T) {
 		{"minimal_verbatim", "minimal", "minimal", true},
 		{"high_verbatim", "high", "high", true},
 		{"medium_maps_to_high", "medium", "high", true},
-		{"off_omitted", "off", "", false},
+		{"off_maps_to_low", "off", "low", true},
 		{"empty_omitted", "", "", false},
 		{"unknown_omitted", "garbage", "", false},
 	}

internal/providers/openai_request.go

@@ -251,14 +251,18 @@ func (p *OpenAIProvider) isGeminiRoute(model string) bool {
 
 // mapGeminiReasoningEffort returns (value, shouldForward). Gemini 3 Preview
 // rejects "medium" with HTTP 400, so we map it to the nearest valid option.
-// "off" and unknown values do not forward — respect user intent (off=disable)
-// and avoid injecting defaults the model might otherwise apply conservatively.
+// "off" maps to "low" (the minimum effort accepted by all Gemini models via
+// OpenAI-compat). Forwarding is required because Gemini's default is "high",
+// which consumes the entire max_tokens budget on reasoning traces and leaves
+// no room for the response. Unknown values do not forward.
 func mapGeminiReasoningEffort(level string) (string, bool) {
 	switch level {
 	case "low", "minimal", "high":
 		return level, true
 	case "medium":
 		return "high", true
+	case "off":
+		return "low", true
 	default:
 		return "", false
 	}