As for the reasoning behind the pose. There are three.
- Any left/right symmetrical pose simplifies modelling time as majority of the model can be built in half and mirrored. Saving asymmetrical details until the end.
- Same goes from the skeletal data, which can be left entirely symmetrical. Which allows for mirroring of animations if needed.
- Finally a pose with separated limbs makes the job of the rigger much easier. Rigging is binding the mesh model to the skeletal data, showing which verts on the model are effected by the various bones. While you could rig a model with it's arms down at it's side it would take a lot more time as a lot of the automatic tools would associate leg and chest verts with arm bones.
As to the T-pose vs. 45-degree inverted Y pose.
When you use skeletal deformation for animation you are using bone positions to create deformations in the shape of the model. This approximates the motion of limbs, but it's only an approximation, and the further from the bind pose you get the more distorted the deformation. There are lots of expensive solutions to fix that, creating target shapes that you blend the verts towards as you use the bones to move the model but those solutions get expensive in memory or cpu time quickly.
So why the "T"? Well if you look at the range of motion a human arm can make it goes from a little past pointing straight up to a little past straight down. This makes out to the side exactly in the middle of the two extreme positions. Since skeletal deformations look worse the further they get from the original pose it makes sense to make the bind pose in the middle.
As FuzzySpoon mentioned the static-T pose has been shifted a bit to arms at more of a 45-degree downward slope. The reason for this is that while horizontal is the middle point for the full range of rotation of the arm, it isn't the middle point for common usage of the arm. Our arms spend most of the time either at our sides or as high as shoulder height. So 45-degrees down is right in the middle of the most common positions for arms in animations. The other issue is that as you lift your arm to horizontal your shoulders tend to rise and clothing bunches up, it's more complicated to model the shoulder area when you have to think of pre-bunched up clothing and raised shoulders.