Thank you for posting your question here.
There is no hard limit regarding the maximum number of images.
More images you have, longer will be the processing time. Hence, the processing strongly depends on the number of images, their resolution, your computer specifications. If your project is very large, it could be better to split into smaller projects and then merge them.
Some users have shared their experience with large projects in this community post: Largest Number of Images.
Regarding your project, it is usually recommended to have 75% frontal and 60% side overlap in general cases.
For glassy buildings, it could be necessary to even increase the overlap since glass is very reflective and therefore the calibration will be harder. Uniform and reflective surface make it difficult for the algorithm to find keypoints.
If your point cloud contains a lot of noise, as it seems according to your attached image, if might be necessary to clean it. You can do it by assign the noisy points to the class "disabled as described here: How to edit the point cloud in the rayCloud.
I am not sure to understand your issue. Is the hill overlapping the building? Maybe there is an issue with the georeferencing? I would need your quality report for a better understanding of your problems.
Additionally, I can see that you were using the Ag multispectral template which is recommended for precision agriculture. According to your screenshot I would recommend you to use 3D Maps or 3D Models templates.
Hope this will help.