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/features/first-run/FirstRunWizard.tsx

@@ -15,11 +15,12 @@
  * `./profile`.
  */
 
-import type { AgentProfile, HttpChatConfig } from "@/domain";
+import type { AgentProfile, HttpChatConfig, OpenCodeConfig } from "@/domain";
 import { Button, IconButton, Input, Panel, Toolbar, cn } from "@/shared";
 import { useFirstRun, type WizardEntry } from "./useFirstRun";
 import {
   defaultHttpChatConfig,
+  defaultOpenCodeConfig,
   parseArgs,
   validateProfile,
   type ProfileErrors,
@@ -190,10 +191,81 @@ function ProfileRow({
       {profile.structuredAdapter === "openAiCompatible" && (
         <HttpChatFields profile={profile} errors={errors} onChange={onChange} />
       )}
+
+      {profile.structuredAdapter === "openCode" && (
+        <OpenCodeFields profile={profile} errors={errors} onChange={onChange} />
+      )}
     </li>
   );
 }
 
+/**
+ * The OpenCode + llama.cpp config section: the base URL of the local
+ * `llama-server` (`/v1`), the model it serves, and an optional API key (the key
+ * itself — a local llama.cpp usually needs none). Rendered only for a `openCode`
+ * profile; edits flow back through `onChange` as a patched `opencode`.
+ */
+function OpenCodeFields({
+  profile,
+  errors,
+  onChange,
+}: {
+  profile: AgentProfile;
+  errors: ProfileErrors;
+  onChange: (p: AgentProfile) => void;
+}) {
+  const opencode = profile.opencode ?? defaultOpenCodeConfig();
+  const patch = (next: Partial<OpenCodeConfig>) =>
+    onChange({ ...profile, opencode: { ...opencode, ...next } });
+
+  return (
+    <fieldset className="mt-1 flex flex-col gap-2 rounded-md border border-border/70 p-2">
+      <legend className="px-1 text-xs font-medium text-muted">
+        Local model (OpenCode + llama.cpp)
+      </legend>
+
+      <label className="flex flex-col gap-1">
+        <Caption>Base URL</Caption>
+        <Input
+          aria-label={`${profile.name} base url`}
+          value={opencode.baseURL}
+          placeholder="http://localhost:8080/v1"
+          invalid={Boolean(errors.baseURL)}
+          onChange={(e) => patch({ baseURL: e.target.value })}
+        />
+        {errors.baseURL && (
+          <small className="text-xs text-danger">{errors.baseURL}</small>
+        )}
+      </label>
+
+      <label className="flex flex-col gap-1">
+        <Caption>Model</Caption>
+        <Input
+          aria-label={`${profile.name} model`}
+          value={opencode.model}
+          placeholder="qwen3-coder-30b"
+          invalid={Boolean(errors.model)}
+          onChange={(e) => patch({ model: e.target.value })}
+        />
+        {errors.model && <small className="text-xs text-danger">{errors.model}</small>}
+      </label>
+
+      <label className="flex flex-col gap-1">
+        <Caption>API key (optional — a local llama.cpp needs none)</Caption>
+        <Input
+          aria-label={`${profile.name} api key`}
+          value={opencode.apiKey ?? ""}
+          placeholder="e.g. sk-no-key"
+          onChange={(e) => {
+            const v = e.target.value;
+            patch({ apiKey: v.length === 0 ? undefined : v });
+          }}
+        />
+      </label>
+    </fieldset>
+  );
+}
+
 /** Parses a number field: blank ⇒ `undefined` (backend applies its default). */
 function parseOptInt(raw: string): number | undefined {
   const trimmed = raw.trim();