Blender model orientation isn’t actually agnostic. If you look at the orthographic views (numpad 7 for top view, numpad 1 for front view, and numpad 3 for right view), you can see that the -Y (negative Y) axis is considered the “front” of the model, and +Z is up. When exporting to glTF, Blender automatically applies a 90 degree rotation, such that the model’s “up” changes from Blender +Z to glTF +Y, and the model’s front changes from Blender’s -Y to glTF’s +Z. STK expects these conventions to be followed, and will adapt the model again to STK’s internal coordinate system.
Rockets are an interesting case for orientation. Often such models are artistically displayed with the tip of the vehicle pointing “up”, as if it were standing on the pad, ready for some kind of vertical launch. But this is NOT what STK wants. STK thinks the direction of travel is the “front” of the vehicle, regardless if it’s flying horizontal or vertical or somewhere in between.
So in Blender, when modeling a rocket, you draw it on its side, as if flying horizontally towards Blender’s “front” -Y axis. When exported to glTF, this becomes glTF’s front axis of +Z, and then when STK imports a glTF, STK adapts it to STK’s front axis of +X, and the rocket flies in that direction. Ideally this should be carried through CZML and fly on this axis in Cesium as well.
One possible wrinkle here, STK allows arbitrary model orientations to be specified in the 3D object properties panel. I don’t know offhand whether these orientations are applied to the CZML or not. If not, that could be the cause of the orientation discrepancy you’re seeing here.
Try modifying the Blender model to fly towards Blender’s -Y axis, and re-export your glTF from there. That should let you clear out any model orientation values you have set in STK.