Hi,
to be honest, the quality of the LIDAR looks fairly bad, actually even worse than the Agisofts dense cloud. So you can't expect that "bad quality + mediocre quality = good quality".
AFAIK, Agisoft uses poisson surface reconstruction for calculating meshes from points. This algorithm requires good normal estimation to achieve nice results. In this case, it seems the normals were not estimated very well, which led to the "bubble" in the front side.
What's the source of the LIDAR point cloud? You could try using Cloudcompare to estimate the normals (it has different settings, so maybe you will be able to achieve better results).
Cheers.