polish(mark): add hybrid action space support to ActionNoiseWrapper

[MarkHolmstrom]

Description

Previously, target policy smoothing could not be used in the TD3 algorithm with hybrid action spaces due to the action object being a dictionary composed of an action_type tensor, an action_args tensor, and a logit tensor instead of the expected action object being the action_type tensor. The PR adds checks for the nested dictionary and adds noise only to the action_args tensor corresponding to the continuous action.

Related Issue

#789

TODO

Check List

• merge the latest version source branch/repo, and resolve all the conflicts
• pass style check
• pass all the tests