Feature Request: allowing NoneType return, support multiple argument for RPC function

[MilkClouds]

I'm trying to set up a clean separation between RL policy and RL environment execution using this project. My basic flow is:

1. Open a new env on remote
2. Run env.step until the episode ends
3. Call env.close

I ran into some issues:

1. No numpy support in zero.
   I solved this by writing a new encoder that handles numpy arrays:

   import asyncio
   import io
   from typing import Any, Type
   
   import numpy as np
   from zero import AsyncZeroClient
   from zero.encoder.generic import GenericEncoder
   from zero.encoder.msgspc import T
   
   class GenericEncoderWithNumpySupport(GenericEncoder):
       def encode(self, data) -> bytes:
           if isinstance(data, np.ndarray):
               buffer = io.BytesIO()
               np.save(buffer, data)
               return super().encode(buffer.getvalue())
           return super().encode(data)
   
       def decode(self, data: bytes) -> Any:
           decoded_data = super().decode(data)
           if decoded_data[1:6] == b"NUMPY":  # MAGIC string for numpy array
               buffer = io.BytesIO(decoded_data)
               return np.load(buffer)
           return decoded_data
   
       def decode_type(self, data: bytes, typ: Type[T]) -> T:
           if issubclass(typ, np.ndarray):
               decoded_data = self.decode(data)
               buffer = io.BytesIO(decoded_data)
               return np.load(buffer)
           return super().decode_type(data, typ)
   
       def is_allowed_type(self, typ: Type) -> bool:
           return super().is_allowed_type(typ) or typ is np.ndarray
   

2. NoneType return disallowed.
   It's an annoying restriction—I just worked around this by returning a dummy value.

3. No support for multiple arguments.
   This was a deal breaker for me. Not supporting multiple arguments made it pretty hard to proceed, so I stopped here.

Would be nice if these issues could get some attention. The numpy part is easily fixable, but the lack of multi-argument support is really limiting.

[Ananto30]

Hi @MilkClouds Nice to see that you are putting some valuable feedback and suggestions, thanks for that.

1. Supporting numpy is a custom choice, and you already made a customer encoder, this should be good.
2. Returning None is by default anti-pattern (it's kind of opinionated). You could return "ok".
3. This is something that needs attention, my initial thought was anything can be sent in a single argument by using pydantic or msgspec. Think of it like a request body. But I am also thinking now we could support multiple arguments, I will take a look.