If you open a factory default Blender and Align Active Camera to View then you can use the camera’s n-panel to see the default viewpoint’s transforms, including the Y rotation of 0.000007°. I recently saw a post or a vid or something that said this was being corrected to 0.000000°.
Looks like it hasn’t been as of the latest 4.3.0 alpha. Neither googling nor asking the duck has turned up any reference to it. Did I dream this?
The issue is not related to Blender, it is just that the average CPU is actually surprisingly bad at math (because you run into the same thing with other programs that do not round values to the first few digits by default).
You can technically force a CPU to perform exact math, but you will lose performance as well (and a DCC app. is a case where you can never have enough of that).
Even if talking about floating point precision/numbers: especially the representation with a significand, exponent and base clearly can “produce” zero. And most 3D modeling apps use the default IEEE_754
even wired in hardware because FPU’s are included in CPU’s since the successor of the Motorola 68030 CPU ( Motorola 68881/2 FPU ) or Intel 8086 CPU ( Intel 8087 FPU ) from the eighties… !!
( Also in IEEE 754 there are actually two versions of zero. )
historical nit pick, while the 8087 was indeed a thing, this was a separate (and expensive [1]) chip you had to buy and install, so it was very much not included, this lasted until the 80486 cpu in 1989 where some models had it on die in but not all (486sx didn’t and you still had to buy 80487SX if you needed the fpu), it wasn’t until the pentium in 1993 they got rid of the concept of a physically separate co-processor.
[1] the boxed version for the 387SX went for $506 1987 dollars (~1400 USD today when adjusted for inflation) so it wasn’t a very popular or even common thing to encounter.
your point still stands though, fpu’s has been built-in on intel since (lets round it up to) the mid 90’s so it’s “been a bit” and most code these days is using IEEE_754, so they can express exact zero (and since the sign is a separate bit also exact -0 which to keep things fun are considered not equal)