[Enhance] Support dumping datasets to pkl file

[shufanwu]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Support dumping datasets to pkl file https://github.com/open-mmlab/mmengine/issues/896

Modification

add load and dump function for data_list in base_dataset.py

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repos?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[HAOCHENYE]

Thanks for your contribution and sorry for the late response. I think that the dump and load should be used like this:

# prepare dataset
dataset = XXXDataset(...)
ori_dataste.dump('saved.pkl')

new_dataset = Dataset.load('saved.pkl')

It means that the dump method should dump all data information, including data_list(unserialized) and data_meta to the file, and load should be a classmethod, which can load the file and recover the original dataset.

[HAOCHENYE]

Hi, thanks for your contribution. The lint seems failed. BTW, I'll test load and dump in some real dataset ASAP 👍