enchance obs manager

[vybhav-ibr]

modify observation manager to return each component of the observations as a separate box, make obs_size calculation compatible with multi-dimensional arrays to support sensors

makes the observation_space of the type collection.OrderedDict, This makes it possible for the policy to segregate the observations into its individual components(this is only possible with skrl, and skrl has a built in function for this called space_to_space), after the observation is divided into its constituents, different NNs can be used for the different observations, for instance the sensor observations like lidar, camera can be encoded into lower dimensional vectors using a CNN, which also helps preserve the spatial relation between the pixels/points. This is pretty much the only way sensor data can be used for RL, because using a fully connected network for the sensor observations will make the network way too large(a few million parameters).

here is an example of how this is used in SKRL code

[jgillick]

This PR is not ready for review:

• You created a new script, in the root directory, instead of actually updating the observation manager. This makes it hard to easily see which parts are different and how to use them. Is this intended to be a new manager?
• There's no example which shows how this is used. Is this for an actual robot project you're working?

Please avoid submitting incomplete PRs