[GraphBolt] There should be an easy way to use fp16 or bf16

## 🔨Work Item

**IMPORTANT:**
* This template is only for dev team to track project progress. For feature request or bug report, please use the corresponding issue templates.
* DO NOT create a new work item if the purpose is to fix an existing issue or feature request. We will directly use the issue in the project tracker.

Project tracker: https://github.com/orgs/dmlc/projects/2

## Description

In our examples, there should be an easy way to switch to lower precision for feature storage and/or training.

Let's say we load a dataset and the memory consumption is high. The user may want to cast the features tensor to float16 or bfloat16. However, it is not easy to do that with our existing abstractions. Either the `Feature` or `TorchBasedFeature` class or the `BuiltinDataset` should have a way to specify the feature datatype either during loading or cast after loading.