feat: send data between agent and worker through files instead of stream-based requests

[aneojgurhem]

This PR changes the way data should be sent between Worker and Scheduling Agent to use files in a shared directory instead of gRPC streams.

Rationale:
Rework communications between scheduling agent and worker so that there is no streams and pass data through file in a shared volume. This will allow some nice optimisations such as:

• Less gRPC communications between scheduling agent and worker
• Simpler RPC methods between scheduling agent and worker
• Tasks that start faster: no need to receive the input payload and dependencies, they will be available in the shared volume at the start of the task due to prefetching
• Data can be consummed when needed during the task; not forcefully received at the start
• A node-wide caching system could be implemented through a volume shared between all the scheduling agents and workers that would allow the scheduling agent to know that some data are already there and reuse them to avoid unnecessary data download. This should be implemented in ArmoniK.Core and should not impact APIs and Worker behavior

What changed ?

• Worker Process RPC is now an Unitary Call (no streams) with a data_location folder in which Payload and Data Dependencies are located as file whose name is the ID of the input
• Agent Get*Data are now Unitary Calls too : The call returns when expected data are located in the data_location folder as file whose name is the ID of the required data
• Agent UploadResultData and SendResult became NotifyResultData. Results should be stored as a file whose name is the ID of the results in the data_location folder then the NotifyResultData method should be called to notify Agent that data were submitted.