Xpress: empty bound check not working -- fixed

[Tahira-2]

Title: Xpress: handle empty integer bounds (fixes #5085)

Problem
The MathOpt parametrised test FractionalBoundsContainNoInteger skipped Xpress because integer variables with unrounded bounds like [0.5, 0.6] are rejected by Xpress at variable creation: Xpress rounds the bounds on input, sees [1, 0], and refuses to add the column. SCIP, GLPK, and HiGHS instead return TerminationReason::kInfeasible.

Fix
Mirrors GLPK's EmptyIntegerBoundsResult() pattern:

In AddNewVariables, detect integer variables where lb <= ub && ceil(lb) > floor(ub). For those, substitute [0, 0] in the bounds passed to Xpress (so it accepts the column) and remember the original bounds in a new empty_integer_bounds_vars_ member.
In Solve(), after the existing ListInvertedBounds check, short-circuit with ResultForIntegerInfeasible(...) if any such variable was recorded — without invoking the optimizer. The original lb/ub flow into the user-facing detail message.
Remove the GTEST_SKIP for kXpress in the test.
Files touched
ortools/math_opt/solvers/xpress_solver.{h,cc} — detection + early-return
ortools/math_opt/solvers/BUILD.bazel — added //ortools/math_opt/core:empty_bounds dep
ortools/math_opt/solver_tests/mip_tests.cc — re-enable the test for Xpress
CMake's solver target is built via a recursive glob, so no CMake change is needed.

Test plan
CI: compile + run all non-Xpress solver tests (covered automatically)
Local Xpress run of FractionalBoundsContainNoInteger — needs an Xpress license; cc @Mizux for verification