I would try few things:
-take photos from more stations(and less photos on each station(rotate camera at about 30 or 45 degree), this could help metashape to have data from more angles with less "jumps" instead of few angles and big jumps.
- if there is lot of traffic and people, wait a little bit between taking photos so the traffic and people completelly changes and metashape evaluate them as invalid feature points.
-taking photos in 360 degree can be useful in your project, but I would also try taking photos from sea line on beach towards buildings, proceed in strips. Buildings gives you plenty of stationary points and you don't need to care about sea and you can control to not have sky in camera view. This way you could be successful with beach surface. For roads and sidewalks can be used another similar technique and you can always wait for clean view.
-I always plane the route where I will be taking photos, so I will ended up with not so much nor few photos which are continuously connected.
- automatically disabled photos after quality estimation could be very missing in some area...I would manually disable only blured images with low exposure time and keep those which have maybe only blured corners, but useful center of image.
Some questions:
-how many photos are not aligned?
-misaligned photos are evenly spreaded acros whole area or are you missing some smaller area completelly and rest is fine?
- do you need detailed pointcloud for your project, or it is enough to recognize objects with certain precision. Do you need all objects, or you can omit some small objects like trash can, trees crown, benches,...?
I'm curious how you will manage all this, as I only do projects on my feet without using such a tall stick