Fix bugs in msr_sqa dataset

[Timothyxxx]

The last version has many problems,

1. Errors in table load-in. Split by a single comma instead of using pandas is wrong.
2. id reduplicated in _generate_examples function.
3. Missing information of history questions which make it hard to use.

I fix it refer to https://github.com/HKUNLP/UnifiedSKG. And we test it to perform normally.

[Timothyxxx]

It shows below when I run test:

FAILED tests/test_dataset_common.py::LocalDatasetTest::test_load_dataset_all_configs_msr_sqa - ValueError: Unknown split "validation". Should be one of ['train', 'test'].

It make no sense for me😂.

[mariosasko]

@albertvillanova Does this PR has some additional fixes compared to #3771 or we can close it?

[albertvillanova]

@mariosasko besides the fix of the DuplicatedKeysError, this PR:

• changes the reading of one of the files: use pandas instead of splitting by comma
• changes the splits: modifying train and adding validation
• adds some extra logic in the processing of the data: adding a new field "question_and_history"

We should decide whether validating these additional changes.

• for example, if we accept as pertinent the addition of the field "question_and_history", this should be added as feature to the info, and the matadata should be regenerated...

[Timothyxxx]

Hi guys, anything we can do to fix that bug👀? @mariosasko @albertvillanova @lhoestq

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[albertvillanova]

@Timothyxxx, thanks for your contribution fixing the bugs. And sorry for the late response.