feat(gate): route prompt-selection through mcpd's server Llm (credential tiering)

Cloud/server keys belong at the k8s/mcpd level; mcplocal handles only the user's
personal tokens. The gate's prompt-selection used mcplocal's LOCAL provider
registry (heavy=anthropic = the user's OAuth token, which is API-gated and 404s),
so selection always degraded.

Now LlmPromptSelector tries sources in priority order: (1) the project's server
Llm via mcpd's inference proxy (POST /api/v1/llms/:name/infer) — cloud/server
keys stay at k8s — then (2) the local personal-token provider as fallback. First
source that returns valid selection JSON wins. Extracted pickCompletionText
(content ?? reasoning_content) + extractSelection helpers.

Wiring: GatePluginConfig.llmProvider (from the project), threaded via
createDefaultPlugin; handleBeginSession builds serverInfer from ctx.postToMcpd.
Verified live: POST .../llms/vllm-current/infer returns valid selection JSON
(qwen3-thinking, 4000-token budget). + selector tests (server preferred, local
fallback on error, server-only). mcplocal suite green (755).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/mcplocal/tests/llm-selector.test.ts

@@ -109,12 +109,49 @@ describe('LlmPromptSelector', () => {
     const selector = new LlmPromptSelector(registry);
     const ac = new AbortController();
 
-    await selector.selectPrompts(['test'], sampleIndex, undefined, ac.signal);
+    await selector.selectPrompts(['test'], sampleIndex, { signal: ac.signal });
 
     const call = (provider.complete as ReturnType<typeof vi.fn>).mock.calls[0]![0] as CompletionOptions;
     expect(call.signal).toBe(ac.signal);
   });
 
+  it('prefers the mcpd server Llm (serverInfer) over the local provider', async () => {
+    const provider = makeMockProvider('{ "selectedNames": ["mqtt-config"], "reasoning": "local" }');
+    const registry = makeRegistry(provider);
+    const selector = new LlmPromptSelector(registry);
+    const serverInfer = vi.fn().mockResolvedValue('{ "selectedNames": ["zigbee-pairing"], "reasoning": "server" }');
+
+    const result = await selector.selectPrompts(['x'], sampleIndex, { serverInfer });
+
+    expect(serverInfer).toHaveBeenCalledOnce();
+    expect(provider.complete).not.toHaveBeenCalled(); // server succeeded → no local call
+    expect(result.selectedNames).toContain('zigbee-pairing');
+    expect(result.reasoning).toBe('server');
+  });
+
+  it('falls back to the local provider when serverInfer throws', async () => {
+    const provider = makeMockProvider('{ "selectedNames": ["mqtt-config"], "reasoning": "local" }');
+    const registry = makeRegistry(provider);
+    const selector = new LlmPromptSelector(registry);
+    const serverInfer = vi.fn().mockRejectedValue(new Error('mcpd HTTP 503'));
+
+    const result = await selector.selectPrompts(['x'], sampleIndex, { serverInfer });
+
+    expect(serverInfer).toHaveBeenCalledOnce();
+    expect(provider.complete).toHaveBeenCalledOnce();
+    expect(result.selectedNames).toContain('mqtt-config');
+    expect(result.reasoning).toBe('local');
+  });
+
+  it('works server-only (null registry) when a serverInfer is supplied', async () => {
+    const selector = new LlmPromptSelector(null);
+    const serverInfer = vi.fn().mockResolvedValue('{ "selectedNames": ["mqtt-config"], "reasoning": "s" }');
+
+    const result = await selector.selectPrompts(['x'], sampleIndex, { serverInfer });
+
+    expect(result.selectedNames).toContain('mqtt-config');
+  });
+
   it('throws when no heavy provider is available', async () => {
     const registry = new ProviderRegistry(); // Empty registry
     const selector = new LlmPromptSelector(registry);