No research here, just a small amount of experience to chip in. :) And no answers, just observations...
First, they're coping with many changes that roll in on the news daily or weekly, many of them more of an upheaval than taking the same number of instructional hours and cutting it into smaller blocks. I don't think you should decide based on unfamiliarity in COVID times.
In my experience, your intuition is right: 90 minutes is a long time to be continuously focused on Zoom, but 45 minutes is very doable. Worse, many people (try to) multitask during remote learning, and 90 minutes is a long time to be dividing attention.
On the other hand, I think a longer break is a good idea. Students seem to find it hard to really break on Zoom. They turn off their mic and video or run to the washroom, but it's hard to do the necessary dissociation: standing up, moving their legs, bustling and talking and sharing impressions with peers, getting a breath of fresh air. If they stare at a screen during break, they won't come back with a fully refreshed ability to sustain attention for another block. For this reason I suspect that 30 minutes of break with explicit reminders to get up and away from the computer is more useful than 15 minutes. But 15 is better than 10 and 10 is better than 5.
As for coming back late and wasting time: not your problem. It's no different from deciding to get a snack and wander luxuriously when learning in person. If they're not back on time, the lesson doesn't stop for them.
How about the problem of momentum? Four blocks with three breaks between them is pretty significant as far as stopping and starting. You would need to be very much on point and purposeful to avoid transition/inertia eating into the 45 minutes from both sides.
One idea is to designate each 45-minute block for a different purpose. For example, if all four are direct instruction, the repeated pattern will accentuate the dreaded feeling of sameness. "Is this the third block or the fourth? Did he teach us that earlier today or last week?" Whereas if you teach the first 45-minute block, give them exercises the second one, teach the third, and give them exercises the fourth, that's quite a nice rhythm.
The same strategy also helps mitigate a 90-minute block. Being present and working while the instructor supervises / is present for questions is not taxing; what's taxing is being attentive while the instructor talks and shows slides (on a screen you may not be viewing or have snapped to half of the screen).