LOS has always worked that way. It's a quick and simple design and it seems to be fine for the majority of uses, since it's worked like that for years and most people seem fine with that.
Also, testing line of sight to any point on the collision mask isn't necessarily desirable. In many cases you don't want to count line of sight if only a tiny fraction of the collision mask is exposed. For example if only a tiny hair-width of the player is showing past a wall, should enemies count that as visible? If it works by any point on the collision polygon, the answer is yes, and then you get enemies that leap in to action chasing you down when the player is still basically hidden (and based on past experience people will probably file this as bugs thinking Construct is calculating this wrong). This then also breaks the symmetry of line-of-sight, meaning the enemies have line of sight to the player, but the player doesn't have line of sight to the enemies. You could also try to fix that by allowing any part of the player collision polygon to have line-of-sight to any part of the enemy collision polygon, but then you get even more absurd results, like a hairline edge of the player's shoulder being able to "see" a hairline edge of an enemy's shoulder counting as both objects having line-of-sight to each other. So then you could try to fix that by having some kind of way to configure the proportion of the area of the collision mask that counts as visible... and now you have a complicated feature that's hard to visualise, difficult for beginners to understand, and tricky to program.
As with most features if you run all the way down the rabbit hole of the various details and edge cases, you end up with lots of complications, which in turn makes Construct more complex and difficult to understand. The main point of the product is to be simple and intuitive for beginners, so we avoid doing this wherever possible. As it stands, you can test line-of-sight to any position, so you could set up a few key image points around the edges of the object, and count it as visible if you have line of sight to any of those points. That's pretty straightforward, doesn't require any new features, and still more or less solves the problem.