It makes a lot of sense because this library is vastly different from usual slippy maps — fluid continuous zoom & rotation, extremely powerful visualization and interactivity potential, no DOM constraints — fitting it into the Leaflet API which was designed for simple image tile-based maps would be very limiting.
On the other hand, Mapbox GL JS is a very complicated technology under the hood, it is much heavier, with more code, lots of algorithms, serious backend technology to support it etc. So there's a place for both libraries — Leaflet remaining the go-to super-simple, friendly and lightweight library for most maps, while GL JS being an experimental ground for maps of the future with unlimited potential.
Also, Mapbox GL JS + Leaflet integration plugin is coming soon. Stay tuned :)
I think that's a mistake — Leaflet is solid and is the foundation on which a lot of extensions are built, to the point where my impression is that it's pretty much the de facto JS-based map widget these days.
While the new library looks decent, I don't see the point of creating a brand new library rather than extending Leaflet, other than to force clients to couple tightly to Mapbox's API, tile server etc.