Based on the team discussion on Sep 16, with the understanding that we are still in the beta phase (0.x) we would like to encourage users to access cuda.core directly:

• easier to remember and type
• easier to teach and communicate
• aligns with the future cuda.core 1.0 plan

This can be done without introducing breaking changes, by moving all internal modules to under cuda.core, and then re-importing them back to cuda.core.experimental. We should find a way to attach a deprecation warning message to each of the re-imported object, without raising warnings when importing cuda.core.experimental. This root module can be kept for other purposes.