[Fixes #65] Transform input point cloud to base_frame to support tilted LiDAR

[abudori3939]

Overview

This pull request resolves the issue reported in #65, where using a physically tilted LiDAR resulted in an incorrect pose estimate. The root cause was that the input point cloud was passed to the registration algorithm before being transformed into the robot's base frame.

Implementation Details

This PR introduces a new preprocessing step in the cloudReceived function:

1. It checks if the header.frame_id of the incoming PointCloud2 message is different from the base_frame_id specified in the parameters.
2. If they are different, it uses tfbuffer_.lookupTransform to get the transform to the base_frame_id.
3. The transform is converted into an Eigen 4x4 matrix.
4. pcl::transformPointCloud is used to transform the entire point cloud into the base_frame_id coordinate system.
5. This newly transformed point cloud is then used for the subsequent registration (NDT/GICP) process.

This change ensures that localization is always performed consistently relative to the robot's base frame, regardless of the LiDAR's mounting orientation.

Verification

With this fix, I have verified the following:

1. The ground plane is now correctly represented as level in the map frame, even when using a tilted LiDAR.
2. As a result of the level representation, obstacles detected by the LiDAR are now reflected with the correct scale in 2D maps.
3. The TF tree is now correctly constructed, reflecting the proper robot configuration.

[rsasaki0109]

Thanks for the fix! The implementation looks good. LGTM, merged.