Add admin LLM model setting; rename Pi to LLM in the UI; drop table Review column

- New server/llm.js consolidates ModelRuntime + model selection: admin
  setting (app_settings.llm_model) > LLM_MODEL/PREFILL_MODEL env > first
  available; used by both prefill and roast evaluation
- GET/PUT /api/admin/llm lists configured models and stores the choice
  (validated against the list; empty = auto; audited); admin page gains
  an LLM section with a model picker
- All user-facing 'Pi agent' wording is now 'LLM'; no_model error message
  no longer references the pi CLI
- /roasts table: Review column removed (review lives in the detail view)

Co-Authored-By: Claude Fable 5 <[email protected]>

server/llm.js

@@ -0,0 +1,44 @@
+// Shared LLM model runtime + selection for every model-backed feature (URL prefill, roast
+// evaluation). One ModelRuntime per process; which model actually runs is decided here:
+// the admin-chosen setting (app_settings.llm_model, passed in as `preferred`) wins, then the
+// LLM_MODEL / PREFILL_MODEL env overrides, then the first available configured model.
+
+import { ModelRuntime } from "@earendil-works/pi-coding-agent";
+
+let modelRuntimePromise = null;
+export async function getModelRuntime() {
+  if (!modelRuntimePromise) modelRuntimePromise = ModelRuntime.create();
+  return modelRuntimePromise;
+}
+
+export const modelKey = (model) => `${model.provider}:${model.id}`;
+
+/** Every configured model, as stable "provider:id" keys for the admin picker. */
+export async function listAvailableModels() {
+  const runtime = await getModelRuntime();
+  return (await runtime.getAvailable()).map((model) => ({
+    key: modelKey(model),
+    name: model.name ?? model.id,
+    provider: model.provider,
+  }));
+}
+
+/** @param {string|null} preferred "provider:id" from the admin setting, or empty for auto. */
+export async function pickModel(preferred = null) {
+  const runtime = await getModelRuntime();
+  for (const candidate of [preferred, process.env.LLM_MODEL, process.env.PREFILL_MODEL]) {
+    if (!candidate) continue;
+    const sep = candidate.indexOf(":");
+    if (sep === -1) continue;
+    const model = runtime.getModel(candidate.slice(0, sep), candidate.slice(sep + 1));
+    if (model) return model;
+  }
+  return (await runtime.getAvailable())[0];
+}
+
+/** Consistent error for "the server has no usable model" across features. */
+export function noModelError() {
+  const err = new Error("No LLM model is configured on the server.");
+  err.code = "no_model";
+  return err;
+}