Add CudaDevicePlacementMixin class

[gabrielmbmb]

Description

This PR adds the CudaDevicePlacementMixin class, and updates vLLM, TransformersLLM and LlamaCppLLM to use it. The class allows defining a list of cuda_devices (ids of the devices), that then will be used to set the environment variable CUDA_VISIBLE_DEVICES for each process. This way, each LLM will use only the cuda_devices allowing the user to define which CUDA devices should be used for each LLM.

In addition, this mixin has an auto mode, which assigns one of the CUDA devices to each LLM avoiding the overlap i.e. loading more than one model in the same device. It's a bit naive, but we can improve it further in the future to take into account the model size and some other factors.

Apart from that this PR also changes:

• When creating the multiprocessing.Pool, only len(self.dag) number of processes will be created, avoiding creating processes that won't do anything.
• Fix an error in which the information of the batches generated by leaf steps wasn't being serialized by the _BatchManager as we were only serializing when adding a batch and not when registering it too.
• Remove calling unneeded self._stop when registering the stop signal for the pipeline and paves the way for a more grateful stop that will be tackled in another PR.
• Removes unnecessaries Locks that were being used to sync the access the manager.dicts (manager.dict already handles that)