[executorch][serialization] Data serialization interface

Introduce data serialization interface.

Differential Revision: [D65947145](https://our.internmc.facebook.com/intern/diff/D65947145/)

ghstack-source-id: 255575940
Pull Request resolved: #7194

Author: lucylq

exir/_serialize/data_serializer.py

@@ -0,0 +1,35 @@
+from abc import ABC, abstractmethod
+from typing import Dict, List, Tuple, Union
+
+from executorch.exir._serialize._cord import Cord
+
+from executorch.exir.schema import Tensor
+
+# Abstract base class that data serializers should adhere to.
+class DataSerializer(ABC):
+    @abstractmethod
+    def __init__(self) -> None:
+        """
+        This initializer may be overridden in derived classes to hold
+        the data required for serialization, eg. configurations.
+        """
+        pass    
+    
+    @abstractmethod
+    def serialize_tensors(self, 
+        tensors: Dict[str, int],
+        tensor_buffer: List[bytes],
+        tensor_metadata: Dict[str, Tensor],
+    ) -> Union[Cord, bytes, bytearray]:
+        """
+        Serializes a list of tensor metadata and tensors emitted by ExecuTorch
+        into a binary blob.
+        """
+        raise NotImplementedError("serialize_data")
+
+    @abstractmethod
+    def deserialize_tensors(self, blob: Union[Cord, bytes, bytearray]) -> Tuple[Dict[str, int], List[bytes], Dict[str, Tensor]]:
+        """
+        Deserializes a blob into a list of tensor metadata and tensors.
+        """
+        raise NotImplementedError("deserialize_data")