Reading all the other pages would have taken forever, but googling "sharpeye gravity hull" probably would have brought up the answer.
Edited:
so probably when I get moved in and have some free time, I'm going to work on Physghosts again finally. The buffers between cells are 2048 units thick iirc, so that's how far your physgun beam will be able to extend into other cells. It's big enough to not notice it hopefully, but the starship shown in video 3 doesn't fit lengthwise into it. I'm planning to change transitions so it can split contraptions apart when transitioning, but I'm afraid large ships might get permanently warped under certain circumstances... what I might need to do is record the parameters and relative positions of the two props at the time of constraint application, then when I need to ghostify parts of it I'll just need to spawn the ghost at the original position relative to its partner, then apply the constraint, then move it to its place relative to the real prop. Traces will also need to be modified to loop through cell borders with a max of two loops. Firebullets will be modified to loop indefinitely, but at a delay and with a curve so bullets will fall if they don't hit something.
We're at more or less the same stage now. I've got a traceline override mostly done, but I've got a few problems to work with, myself. I came across a similar idea for transitions; I didn't think the split-between-cells idea would work very well, because something that may have been welded with a 8-unit gap now has a (cell-size, so 27,000?) gap between entities. Knowing GMod, it's more likely to freak out than simulate that correctly, so I've left the idea on the bench for now.
I'd like to propose a collaboration on this; not only would it speed development, but problems common to both can be solved. My collision checks and clientside rendering seem to be working pretty well, if not unoptimised; I'm not getting any issues with the flashlight, and everything seems to be colliding correctly. Working on a project identical to yours but with different issues seems counter-productive; but hey, it's your decision.
(On an unrelated note: about the tracelines, you wouldn't necessarily have to limit the number of times the trace goes through - traces don't go on forever, and their end position would be a certain distance away from the starting position. I would probably just call the traceline function recursively until the start position is in the same cell as the end position. Depending on performance constraints, you may want to limit it, but I think the impact of going until the end position has been reached is negligible.)