Fix cvxpy failure in nightly testing (#656)

When building the optimization problem from Papilo, the problem category was set to `LP` regardless of the type of problem. This caused the size of `presolve_data.var_flags` to be zero, leading to a illegal memory access in `set_implied_integers`. 

Closes #559.

Authors:
  - Nicolas L. Guidotti (https://github.com/nguidotti)

Approvers:
  - Hugo Linsenmaier (https://github.com/hlinsen)

URL: #656

Author: nguidotti

cpp/src/mip/presolve/third_party_presolve.cpp

@@ -185,14 +185,17 @@ papilo::Problem<f_t> build_papilo_problem(const optimization_problem_t<i_t, f_t>
 
 template <typename i_t, typename f_t>
 optimization_problem_t<i_t, f_t> build_optimization_problem(
-  papilo::Problem<f_t> const& papilo_problem, raft::handle_t const* handle_ptr)
+  papilo::Problem<f_t> const& papilo_problem,
+  raft::handle_t const* handle_ptr,
+  problem_category_t category)
 {
   raft::common::nvtx::range fun_scope("Build optimization problem");
   optimization_problem_t<i_t, f_t> op_problem(handle_ptr);
 
   auto obj = papilo_problem.getObjective();
   op_problem.set_objective_offset(maximize_ ? -obj.offset : obj.offset);
   op_problem.set_maximize(maximize_);
+  op_problem.set_problem_category(category);
 
   if (papilo_problem.getNRows() == 0 && papilo_problem.getNCols() == 0) {
     // FIXME: Shouldn't need to set offsets
@@ -426,7 +429,7 @@ std::optional<third_party_presolve_result_t<i_t, f_t>> third_party_presolve_t<i_
                  papilo_problem.getConstraintMatrix().getNnz());
 
   auto opt_problem =
-    build_optimization_problem<i_t, f_t>(papilo_problem, op_problem.get_handle_ptr());
+    build_optimization_problem<i_t, f_t>(papilo_problem, op_problem.get_handle_ptr(), category);
   auto col_flags = papilo_problem.getColFlags();
   std::vector<i_t> implied_integer_indices;
   for (size_t i = 0; i < col_flags.size(); i++) {