feat(opencode): remplace le provider Ollama par llama.cpp

Le tool-calling local ne fonctionnait jamais via Ollama. Refonte du
support local d'OpenCode autour de llama.cpp: profil, catalogue,
matérialisation de la config OpenCode et surface first-run alignés sur
llama-server (backend + frontend).

QA vert (commandes réelles): domain 244, application 81+64, infra 263
(10 échecs = bind-port sandbox identiques sur develop, non-régression),
frontend 574, tsc propre. Réserve E2E live non bloquante faute de
llama-server joignable.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

frontend/src/domain/index.ts

@@ -662,10 +662,31 @@ export interface HttpChatConfig {
   maxToolIterations?: number;
 }
 
-/** Configuration for an OpenCode process-backed profile. */
+/**
+ * Configuration for an OpenCode process-backed profile pointed at its own
+ * llama.cpp (OpenAI-compatible) endpoint. Mirror of the backend `OpenCodeConfig`
+ * (camelCase wire format: `baseURL`, `apiKey?`, `model`, `reasoning?`,
+ * `attachment?`). OpenCode is a locally-piloted CLI, distinct from the in-process
+ * {@link HttpChatConfig} adapter.
+ */
 export interface OpenCodeConfig {
-  /** Full OpenCode model id, for example `ollama/qwen3-coder:30b`. */
-  model?: string;
+  /**
+   * OpenAI-compatible base URL served by `llama-server` (http/https), for
+   * example `http://localhost:8080/v1`.
+   */
+  baseURL: string;
+  /**
+   * Optional API key forwarded to the provider — the key *itself*, not an env
+   * var name (a local llama.cpp typically needs none / a placeholder). Omitted
+   * when blank (mirrors the backend `skip_serializing_if`).
+   */
+  apiKey?: string;
+  /** Model name served by `llama-server`, for example `qwen3-coder-30b`. */
+  model: string;
+  /** Enables model reasoning. Effective default: `true`. */
+  reasoning?: boolean;
+  /** Enables attachments. Effective default: `false`. */
+  attachment?: boolean;
 }
 
 /**