feat: add custom OpenAI-compatible provider with custom headers and parameters

- Add Custom provider for any OpenAI-compatible API
- All providers now support custom headers and parameters
- providerSettings stored per-provider with headers and parameters
- Example: disable thinking in Ollama, add tools, etc.

backend/src/services/ai/lmstudio.ts

@@ -1,13 +1,18 @@
 import type { AIProvider, AIProviderConfig, AIProviderResult } from './provider';
+import { ProviderSettings } from './openai';
 
 export class LMStudioProvider implements AIProvider {
   provider = 'lmstudio' as const;
   private baseUrl: string;
   private model: string;
+  private customHeaders?: Record<string, string>;
+  private customParameters?: Record<string, unknown>;
 
-  constructor(config: AIProviderConfig) {
+  constructor(config: AIProviderConfig, settings?: ProviderSettings) {
     this.baseUrl = config.baseUrl || 'http://localhost:1234/v1';
     this.model = config.model || 'local-model';
+    this.customHeaders = settings?.headers;
+    this.customParameters = settings?.parameters;
   }
 
   async generate(prompt: string, systemPrompt?: string, options?: { jsonMode?: boolean }): Promise<AIProviderResult> {
@@ -19,18 +24,26 @@ export class LMStudioProvider implements AIProvider {
     
     messages.push({ role: 'user', content: prompt });
 
-    const requestBody = {
+    const requestBody: Record<string, unknown> = {
       model: this.model,
       messages,
       temperature: 0.7,
       max_tokens: 2000,
+      ...this.customParameters,
+    };
+
+    if (options?.jsonMode) {
+      requestBody.response_format = { type: 'json_object' };
+    }
+
+    const headers: Record<string, string> = {
+      'Content-Type': 'application/json',
+      ...this.customHeaders,
     };
 
     const response = await fetch(`${this.baseUrl}/chat/completions`, {
       method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-      },
+      headers,
       body: JSON.stringify(requestBody),
     });
 
@@ -63,7 +76,9 @@ export class LMStudioProvider implements AIProvider {
 
   async validate(): Promise<boolean> {
     try {
-      const response = await fetch(`${this.baseUrl}/models`);
+      const response = await fetch(`${this.baseUrl}/models`, {
+        headers: this.customHeaders,
+      });
       return response.ok;
     } catch {
       return false;