Fix scan accept/reject gating and pose bookkeeping; expose plane rejection stats

[tejasps28]

This PR fixes a logic bug in LidarSlam.cpp that effectively disabled scan rejection, corrects
last_T_w_lidar bookkeeping so motion deltas are computed against the last accepted pose, and makes
plane rejection diagnostics visible in /super_odometry_stats. It also wires the global
use_imu_roll_pitch parameter so YAML settings are actually respected.

Changes

1. super_odometry/src/LidarProcess/LidarSlam.cpp
   • checkMotionThresholds() now returns the real acceptResult (no forced accept) and guards
     invalid dt.
   • last_T_w_lidar is initialized only on the first frame and updated after accept/reject.
   • Plane rejection breakdown fields (plane_*) are populated from MatchRejectionHistogramPlane.
2. super_odometry/src/parameter/parameter.cpp
   • use_imu_roll_pitch is declared and loaded into global USE_IMU_ROLL_PITCH, with logging.

Why

• Previously, scan rejection never took effect because the code forced acceptance at the end.
• last_T_w_lidar was overwritten every frame, making motion statistics and gating inconsistent.
• Plane correspondence health metrics were always zero, blocking diagnostics.

Validation

• Re-ran humanoid benchmarks (aligned). Metrics such as n_iterations, latency, and translation
  increments remained stable, while plane_* diagnostics are now non-zero and informative.
• No regression observed in map visualization on the test bag (expected if rejection thresholds
  rarely trigger).