refactor(llm-api): implement clean architecture following scraper pattern

Split monolithic 1012-line main.rs into layered hexagonal architecture:
- Domain: entity types and LlmError enum
- Application: prompt building, sampler construction, tool call parsing
- Infrastructure: LlamaEngine wrapping llama-cpp-2 with isolated unsafe transmute
- Presentation: Axum handlers, middleware (auth), error chain, router
- Config: type-safe AppConfig with LazyLock
- Bootstrap: Application struct with build() + run()

Resolves build_sampler/build_sampler_params duplication.
Adds simple web chat UI at GET /.

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

src/presentation/router.rs

@@ -0,0 +1,27 @@
+//! Axum router assembly.
+
+use std::sync::Arc;
+
+use axum::middleware;
+use axum::routing::{get, post};
+use axum::Router;
+use tower_http::cors::CorsLayer;
+
+use super::handler::{chat, chat_ui, health, models};
+use crate::presentation::middleware::auth::auth_middleware;
+use crate::presentation::state::AppState;
+
+/// Build the main application router with all routes and middleware.
+pub fn build_router(state: Arc<AppState>) -> Router {
+    Router::new()
+        // Public routes (no auth)
+        .route("/", get(chat_ui::chat_ui))
+        .route("/health", get(health::health_check))
+        .route("/v1/models", get(models::list_models))
+        // Chat completions (auth-protected)
+        .route("/v1/chat/completions", post(chat::chat_completions))
+        .route_layer(middleware::from_fn(auth_middleware))
+        // Global middleware
+        .layer(CorsLayer::permissive())
+        .with_state(state)
+}