Fix _STEP and _RESET for TempVMPackingEnv (fixes #36)

[joefarrington]

This pull request includes several changes to the TempVMPackingEnv defined in or_gym/envs/classic_or/vmpacking.py that fixes #36

• or_gym/envs/classic_or/vmpacking.py: Renamed step method to _STEP so that customized logic is used. In current implementation steps use the _STEP from the parent class VMPackingEnv.
• or_gym/envs/classic_or/vmpacking.py: Updated pm_state array operations in _STEP to exclude the first element of the demand array and addressed rounding issues by setting values less than self.tol to 0.
• or_gym/envs/classic_or/vmpacking.py: Modified update_state method to use np.float32 for data_center and np.uint8 for action_mask and avail_actions to ensure proper data types. This ensures that the tests pass, because they include a check against the observation_space which uses these types.
• or_gym/envs/classic_or/vmpacking.py: Updated _RESET method to initialize state with dictionaries containing np.uint8 and np.float32 data types for action_mask, avail_actions, and state. _RESET method previously gave an error because the default state was a NumPy array but update_state tries to set values using dictionary keys.

This pull request addresses the default setting where self.mask=True, but does not make edits to ensure that, if the user sets self.mask=False, the state consists of just the NumPy array (vs dictionary containing arrays for action mask, available action, and the state).