The M3 refresh of the Studio and Pro was probably scrapped because of issues with the Ultra that the M4 generation would rectify.
There were rumors that Apple was developing the M3 Ultra as a single-die standalone chip instead of two M3 Maxes stitched together. Indeed the M3 Max is reported to not feature the UltraFusion interconnect that enabled this. The N3B process node the M3 is based on has poor yield, and the bigger the physical size of the die means the bigger a physical area for a lithography defect. This is entirely conjecture but I'm guessing test yields of the M3 Ultra were so poor it wasn't worth Apple's while to release it, and they couldn't just stitch two Max chips together because they removed the UltraFusion connector.
The N3E process node has significantly better yield, making a single-die Ultra chip much more viable.