Because the actions are the part that changes most and a variant per shape would run away from you. The usage frame puts two side by side, drops to one, and swaps the pair for a search field, all by editing the instance. The axes cover the text, which is where the height actually comes from.
Why does it have no fill and no border?
An empty state is nearly always already inside a boundary: a card, a panel, the body of a table. Giving it a surface of its own would draw a box inside a box. The 14px radius is there for the times you do want to paint it.
What is the media component for?
So the block does not need a media axis multiplying its four variants by three. Icon is a 16px Tabler glyph on a 32px base.muted tile, Avatar is a single 48px circle, and AvatarGroup is three 32px circles overlapping by 8 with a 2px white ring each.
Why does only the Icon type get a tile?
An avatar is already a shape with a defined edge, so a muted square behind it reads as a second frame. The group's white rings do the separating instead, which is the same job the tile does for a glyph that would otherwise float.
What happens at Show Title = No and Show Description = No?
You get the 176 tall version: media, then the actions. It is the shape to reach for when the surrounding panel already carries a heading and repeating it inside the block would say the same thing twice.