feat(model-server): frontend modèles locaux — badge de statut & CRUD serveurs (#35) et wizard multi-profils OpenCode (#36)

Sprint « Modeles locaux », couche frontend.

#35 :
- F35.1 badge de statut de lancement du serveur local
  (ModelServerLaunchBadge + useAgentsModelServer).
- F35.2 feature model-servers : CRUD (ModelServersPanel / useModelServers /
  gateway modelServer) et ModelServerSelect.

#36 :
- Liste multi-profils OpenCode dans le wizard de premier lancement,
  gateway de clonage (clone_opencode_profile_from_seed).

Tests verts (exécution réelle) : tsc propre, vitest 608/608.

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

frontend/src/domain/index.ts

@@ -13,11 +13,92 @@ export interface HealthReport {
   note: string | null;
 }
 
+/**
+ * Lifecycle status of a local model server during an agent launch (F35, mirror
+ * of the backend `ModelServerStatusDto`, tagged on `state`, camelCase wire).
+ *
+ * Contract note — the backend event carries the `serverId` on the *envelope*
+ * ({@link DomainEvent} `modelServerStatusChanged`), not inside the status, and it
+ * does NOT emit `baseURL`/`model` in the status payload: the UI reads what the
+ * backend actually sends. `reused` (on `ready`) tells apart a freshly-started
+ * server from a reused running one.
+ */
+export type ModelServerStatus =
+  | { state: "notConfigured" }
+  | { state: "probing" }
+  | { state: "starting" }
+  | { state: "ready"; reused: boolean }
+  | { state: "failed"; code: string; message: string };
+
+/**
+ * Stop policy of a managed local model server (mirror of the backend
+ * `StopPolicyDto`, camelCase wire):
+ * - `keepAlive`: leave the process running,
+ * - `stopOnAppExit`: stop when IdeA exits (the effective default),
+ * - `stopWhenUnused`: stop when no profile references it.
+ */
+export type StopPolicy = "keepAlive" | "stopOnAppExit" | "stopWhenUnused";
+
+/**
+ * A declared local model server (F35, mirror of the backend flat
+ * `LocalModelServerConfigDto`, camelCase wire). Global to IdeA (not project
+ * scoped). Referenced by an OpenCode profile through
+ * `OpenCodeConfig.localModelServerId` (= this `id`).
+ *
+ * Contract notes:
+ * - `id` is a UUID **minted client-side** on create (the backend requires a valid
+ *   UUID; it does not generate the server id, only an internal model id it hides).
+ * - `servedModelName` is what the user configures and what is sent to OpenCode as
+ *   the `model` — the backend's internal `model.id`/`model.label` are not exposed.
+ * - `modelPath` is required by the backend when `autoStart` is `true`.
+ */
+export interface LocalModelServerConfig {
+  id: string;
+  /** Only `llamaCpp` is supported today (camelCase, not "llamacpp"). */
+  kind: "llamaCpp";
+  name: string;
+  /** OpenAI-compatible base URL served by `llama-server` (normalised to `/v1`). */
+  baseURL: string;
+  port: number;
+  /** Absolute `.gguf` path. Required when `autoStart` is `true`. */
+  modelPath?: string;
+  /** Model name exposed by the server and sent to OpenCode as `model`. */
+  servedModelName: string;
+  /** Explicit `llama-server` executable path/command (optional). */
+  binaryPath?: string;
+  args: string[];
+  autoStart: boolean;
+  stopPolicy: StopPolicy;
+}
+
 /** A domain event relayed from the backend (tagged union on `type`). */
 export type DomainEvent =
   | { type: "projectCreated"; projectId: string }
   | { type: "agentLaunched"; agentId: string; sessionId: string }
   | { type: "agentExited"; agentId: string; code: number }
+  | {
+      /**
+       * A local model server changed lifecycle state while (re)launching an
+       * agent whose OpenCode profile binds `localModelServerId` (F35). Keyed by
+       * `serverId`; the frontend correlates it back to the agent through the
+       * profile's `opencode.localModelServerId`.
+       */
+      type: "modelServerStatusChanged";
+      serverId: string;
+      status: ModelServerStatus;
+    }
+  | {
+      /**
+       * An agent launch failed before a runtime session was created (F35). Unlike
+       * a generic "agent failed", this carries an actionable `message` and a
+       * stable `code`; `cause` namespaces the failure (e.g. `"model_server"`).
+       */
+      type: "agentLaunchFailed";
+      agentId: string;
+      cause: string;
+      code: string;
+      message: string;
+    }
   | { type: "agentProfileChanged"; agentId: string; profileId: string }
   | { type: "agentBusyChanged"; agentId: string; busy: boolean }
   | {
@@ -687,6 +768,13 @@ export interface OpenCodeConfig {
   reasoning?: boolean;
   /** Enables attachments. Effective default: `false`. */
   attachment?: boolean;
+  /**
+   * Optional id of a managed local model server (F35) this profile binds to.
+   * When set, IdeA can start/stop the referenced `llama-server` for the profile;
+   * omitted when the profile points at an externally-managed endpoint (mirrors
+   * the backend `skip_serializing_if`). Opaque string id on the wire.
+   */
+  localModelServerId?: string;
 }
 
 /**