feat: auto LLM feedback runner + problem link + 5xx retry

- Add SubmissionFeedbackRunner: async background queue for auto LLM feedback
- Enqueue feedback generation after each submission in submitProblem()
- Register runner in main.cc with CSP_FEEDBACK_AUTO_RUN env var
- Add problem_title to GET /api/v1/submissions/{id} response
- Frontend: clickable problem link on submission detail page
- Enhance LLM prompt with richer analysis dimensions
- Add 5xx/connection error retry (max 5 attempts) in Python LLM script

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

backend/include/csp/services/submission_feedback_runner.h

@@ -0,0 +1,41 @@
+#pragma once
+
+#include "csp/db/sqlite_db.h"
+
+#include <cstddef>
+#include <cstdint>
+#include <deque>
+#include <mutex>
+#include <string>
+
+namespace csp::services {
+
+class SubmissionFeedbackRunner {
+ public:
+  static SubmissionFeedbackRunner& Instance();
+
+  void Configure(std::string db_path);
+
+  /// Enqueue a submission for async feedback generation.
+  bool Enqueue(int64_t submission_id);
+
+  /// Scan DB for submissions without feedback and enqueue them.
+  void AutoStartIfEnabled(db::SqliteDb& db);
+
+  size_t PendingCount() const;
+
+ private:
+  SubmissionFeedbackRunner() = default;
+  void StartWorkerIfNeededLocked();
+  void WorkerLoop();
+  void RecoverPendingLocked();
+
+  std::string db_path_;
+  mutable std::mutex mu_;
+  std::deque<int64_t> queue_;
+  size_t pending_jobs_ = 0;
+  bool worker_running_ = false;
+  bool recovered_ = false;
+};
+
+}  // namespace csp::services