fix build bugs

Dockerfile

@@ -57,7 +57,16 @@ RUN git clone --depth 1 --branch ${LLAMA_CPP_REF} https://github.com/ggerganov/l
 # Build C++ manager (Boost.Beast + JSON)
 COPY cpp/ /build/cpp/
 RUN g++ -std=c++17 -O2 -DBOOST_ERROR_CODE_HEADER_ONLY -pthread \
-    /build/cpp/*.cpp -o /build/llm-manager
+    -I/build/cpp \
+    /build/cpp/config.cpp \
+    /build/cpp/http_helpers.cpp \
+    /build/cpp/llm_manager.cpp \
+    /build/cpp/llm_manager_types.cpp \
+    /build/cpp/model_manager.cpp \
+    /build/cpp/request_parsing.cpp \
+    /build/cpp/runtime_components.cpp \
+    /build/cpp/server.cpp \
+    -o /build/llm-manager
 
 # Runtime stage
 FROM debian:bookworm-slim

cpp/llm_manager_types.h

@@ -147,7 +147,7 @@ struct RequestContext {
     std::atomic<RequestState> state{RequestState::QUEUED};
     std::atomic<bool> cancelled{false};
     std::chrono::steady_clock::time_point created_at{std::chrono::steady_clock::now()};
-    std::chrono::steady_clock::time_point enqueue_time{created_at};
+    std::chrono::steady_clock::time_point enqueue_time{std::chrono::steady_clock::now()};
     std::chrono::steady_clock::time_point start_time{};
     std::mutex mu;
     std::condition_variable cv;

cpp/runtime_components.cpp

@@ -6,6 +6,7 @@
 
 #include <algorithm>
 #include <sstream>
+#include <utility>
 
 RateLimiterStore::RateLimiterStore(const RateLimitConfig &config)
     : requests_per_minute_(std::max(0, config.requests_per_minute)),

cpp/server.cpp

@@ -9,6 +9,7 @@
 
 #include <algorithm>
 #include <atomic>
+#include <utility>
 
 namespace beast = boost::beast;
 namespace http = beast::http;