fix: stale modelSlug (#100)

1. fix: remove definition of `currentConversation` as it is not used (to
fix the following error when running `npm run build:prd:chrome`
```bash
> webapp@0.0.0 build:prd:chrome
> bash -c 'source ./scripts/build-prd-chrome.sh'

🧩 Start building Chrome Extension ...
📦 Version: 2.16.0
📦 Monorepo Revision: 9fc435
📦 Beta Build: false
📦 API Endpoint: https://app.paperdebugger.com

❌ Failed to build Chrome Extension, please check logs/build.log

> webapp@0.0.0 build
> tsc -b && npm run _build:default && npm run _build:background && npm run _build:intermediate && npm run _build:settings && npm run _build:popup

src/views/settings/setting-text-input.tsx(32,13): error TS6133: 'currentConversation' is declared but its value is never read.
```
2. fix: change matching from substring to exact, to prevent
`gpt-4.1-mini` from matching to `gpt-4.0` (edit: typo, should be
`gpt-4.1`), making the method more robust to the order of supported
models defined.

3. [edit] refactor: use IsCustom and not redefine logic in NewAIClientV2

Author: wjiayis

internal/services/toolkit/client/client_v2.go

@@ -63,21 +63,33 @@ func NewAIClientV2(
 	logger *logger.Logger,
 ) *AIClientV2 {
 	database := db.Database("paperdebugger")
+	
+	llmProvider := &models.LLMProviderConfig{
+		APIKey: cfg.OpenAIAPIKey,
+	}
+
+	var baseUrl string
+	var apiKey string
+	var modelSlug string
 
-	if cfg.InferenceBaseURL != "" && cfg.InferenceAPIKey != "" {
-		oaiClient := openai.NewClient(
-			option.WithBaseURL(cfg.InferenceBaseURL+"/openrouter"),
-			option.WithAPIKey(cfg.InferenceAPIKey),
-		)
-		CheckOpenAIWorksV2(oaiClient, cfg.InferenceBaseURL+"/openrouter", "openai/gpt-5-nano", logger)
+	// User specified their own API key, use the OpenAI-compatible endpoint
+	if llmProvider != nil && llmProvider.IsCustom() {
+		baseUrl = cfg.OpenAIBaseURL
+		apiKey = cfg.OpenAIAPIKey
+		modelSlug = "gpt-5-nano"
+	// Use the default inference endpoint
 	} else {
-		oaiClient := openai.NewClient(
-			option.WithBaseURL(cfg.OpenAIBaseURL),
-			option.WithAPIKey(cfg.OpenAIAPIKey),
-		)
-		CheckOpenAIWorksV2(oaiClient, cfg.OpenAIBaseURL, "gpt-5-nano", logger)
+		baseUrl = cfg.InferenceBaseURL + "/openrouter"
+		apiKey = cfg.InferenceAPIKey
+		modelSlug = "openai/gpt-5-nano"
 	}
 
+	oaiClient := openai.NewClient(
+		option.WithBaseURL(baseUrl),
+		option.WithAPIKey(apiKey),
+	)
+	CheckOpenAIWorksV2(oaiClient, baseUrl, modelSlug, logger)
+
 	toolRegistry := initializeToolkitV2(db, projectService, cfg, logger)
 	toolCallHandler := handler.NewToolCallHandlerV2(toolRegistry)
 

webapp/_webapp/src/views/settings/setting-text-input.tsx

@@ -29,7 +29,7 @@ export function createSettingsTextInput<K extends SettingKey>(settingKey: K) {
     password = false,
   }: SettingsTextInputProps) {
     const { settings, isUpdating, updateSettings } = useSettingStore();
-    const { currentConversation, setCurrentConversation } = useConversationStore();
+    const { setCurrentConversation } = useConversationStore();
     const [value, setValue] = useState<string>("");
     const [originalValue, setOriginalValue] = useState<string>("");
     const [isEditing, setIsEditing] = useState<boolean>(false);
@@ -59,7 +59,7 @@ export function createSettingsTextInput<K extends SettingKey>(settingKey: K) {
           // try to find a model that matches the current slug
           const currentSlugLower = latest.modelSlug.toLowerCase();
           const matchingModel = response.models.find(m =>
-            currentSlugLower.includes(m.name.toLowerCase())
+            currentSlugLower === m.name.toLowerCase()
           );
           // fall back to the first model in the list
           const newSlug = matchingModel?.slug ?? response.models[0].slug;