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/domain/error.rs

@@ -0,0 +1,38 @@
+//! Domain-level error types.
+//!
+//! Framework-agnostic errors that can be mapped to HTTP errors
+//! at the presentation layer.
+
+use thiserror::Error;
+
+/// Errors originating from LLM inference operations.
+#[derive(Error, Debug)]
+pub enum LlmError {
+    /// Invalid request parameters
+    #[error("Invalid request: {0}")]
+    InvalidRequest(String),
+
+    /// Model or inference error
+    #[error("Model error: {0}")]
+    Model(String),
+
+    /// Authentication failure
+    #[error("Authentication failed")]
+    Unauthorized,
+
+    /// Internal/unexpected error
+    #[error("Internal error: {0}")]
+    Internal(String),
+}
+
+impl From<String> for LlmError {
+    fn from(s: String) -> Self {
+        LlmError::Internal(s)
+    }
+}
+
+impl From<&str> for LlmError {
+    fn from(s: &str) -> Self {
+        LlmError::Internal(s.to_string())
+    }
+}