1) There's no "best", it just depends on what is the more practical for you as a maker.
For the end-user it changes about nothing.
2) It depends on the type of game, on the resolution of the game, on what's more practical for you and for the end-user in matter of download-size.
3) If the obstacle is not supposed to move, I guess you can use a single sprite which displays the image.
In the end, what matters in collision is the collision polygon. Hotspot don't exist in C2.
Using an invisible sprite/box over another textured sprite can be useful if you want several collision areas on your texture (since a collision polygon is limited to 8 points anyway).
So in the end, it depends on the game, the type of obstacle and the intent of the maker.
4) It is just what's more practical for you and your coding.
It also depends on the game you're making.
I strongly suggest you to read Ashley's blog article about optimization, it contains what really makes a difference.
In the end the "what is best" type of question is often answered by "it depends on your taste and abilities", what's more convenient to you as a maker and if it fits in the logic of the game you're currently making.