Inconsistent/inaccurate names

[laurenyu]

Is your feature request related to a problem? Please describe.
Some of our naming choices could've been better, and this can lead to confusion over what a class/parameter/variable is supposed to do. (I'm definitely very guilty of this.)

Describe the solution you'd like
Here are the ones I'm thinking we should rename as part of #1459:

Current	Proposed	Notes
s3_input	TrainingInput	s3_input was created back in 2017 when SageMaker had only training jobs and endpoints. Now we have many types of inputs that can be in S3, so it would be good to match s3_input to our newer classes. (Also, it doesn’t follow Python convention for a class name.)
distributions	distribution	The parameter describes how to handle distributed training with certain estimators, and was honestly just a typo on my part.
train_use_spot_images	use_spot_images	The "train" is redundant on estimator parameters. This also applies to the other parameters that start with "train." (Also, the more accurate adjective would be "training.")
image_name, image	image_uri	"URI" is more accurate than "name", and I think it's nice to have the precision of specifying "URI" over just calling it "image" since "image" alone could suggest some kind of object that has more info.
RealTimePredictor	Predictor	The class was originally written thinking there would be some kind of "BatchPredictor." Instead, we went with the entirely separate Transformer.
endpoint	endpoint_name	RealTimePredictor is the only predictor class in this SDK that doesn't call it endpoint_name.
session	sagemaker_session	The S3 utility classes are inconsistent with the other classes on this one, and I'm not sure they need to be.