Add support for Amazon S3 URI in predictor serialization

[juliangrueber]

Description

The serialisation of the predictor expects a Path from Pathlib:

gluonts/src/gluonts/model/predictor.py

Lines 78 to 80 in b63fc05

	def serialize(self, path: Path) -> None:
	# serialize Predictor type
	with (path / "gluonts-config.json").open("w") as fp:

Passing an S3 URI causes the method to fail. The reason is the usage of Pathlib

from pathlib import Path
Path("s3://my-bucket") / "my_file"

results in PosixPath('s3:/my-bucket/my_file')

I'm using AutuGluon on SageMaker Studio where I am limited in disk size and I want to track all my experiments on Amazon S3. Storing everything on S3, I easily can access the Tensorflow board without using SageMaker Studio. Therefore, it would be great if this method would also support S3 URIs.

CC: @shchur