feat: multi-provider support — Anthropic + Ollama (P2-002)

Add ProviderService wrapping Pi SDK's ModelRegistry for multi-provider
LLM support. Built-in providers (Anthropic, OpenAI, Google, xAI, etc.)
auto-discovered; Ollama registered via OLLAMA_BASE_URL env var;
custom providers via MOSAIC_CUSTOM_PROVIDERS JSON env var.

- ProviderService: wraps ModelRegistry, manages provider lifecycle
- ProvidersController: GET /api/providers, GET /api/providers/models
- AgentService: accepts provider/model params on session creation
- ChatGateway: passes optional provider/modelId from chat messages
- @mosaic/types: new provider/model type definitions

Closes #20

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

packages/types/src/index.ts

@@ -2,3 +2,4 @@ export const VERSION = '0.0.0';
 
 export * from './chat/index.js';
 export * from './agent/index.js';
+export * from './provider/index.js';

packages/types/src/provider/index.ts

@@ -0,0 +1,54 @@
+/** Known built-in LLM provider identifiers */
+export type KnownProvider =
+  | 'anthropic'
+  | 'openai'
+  | 'google'
+  | 'ollama'
+  | 'xai'
+  | 'groq'
+  | 'openrouter'
+  | 'zai'
+  | 'mistral';
+
+/** Provider identifier — known providers or custom string */
+export type ProviderId = KnownProvider | string;
+
+/** Describes an available LLM model */
+export interface ModelInfo {
+  id: string;
+  provider: ProviderId;
+  name: string;
+  reasoning: boolean;
+  contextWindow: number;
+  maxTokens: number;
+  inputTypes: ('text' | 'image')[];
+  cost: {
+    input: number;
+    output: number;
+    cacheRead: number;
+    cacheWrite: number;
+  };
+}
+
+/** Describes an available provider */
+export interface ProviderInfo {
+  id: ProviderId;
+  name: string;
+  available: boolean;
+  models: ModelInfo[];
+}
+
+/** Configuration for a custom (non-built-in) provider */
+export interface CustomProviderConfig {
+  id: string;
+  name: string;
+  baseUrl: string;
+  apiKey?: string;
+  models: Array<{
+    id: string;
+    name: string;
+    reasoning?: boolean;
+    contextWindow?: number;
+    maxTokens?: number;
+  }>;
+}