Split bandit_feedback to train and test when is_timeseries_split=True

[nomuramasahir0]

Overview

This PR modifies the output of the obtain_batch_bandit_feedback function in obp/dataset/real.py.
In the current implementation, even if is_timeseries_split=True, one dict containing the key such as ..._test was returned.
In this PR, if is_timeseries_split=True, split to create two dict and return them to users.

Note

The modifications due to the changes in this PR are limited to the scope of test module.
Therefore, for example, for the following file, it is expected that execution will result in an error due to this PR.

zr-obp/benchmark/ope/train_regression_model.py

Line 48 in c870a20

if is_timeseries_split:

[usaito]

@nmasahiro Thanks! Can you address the following two points?

• I think n_rounds key is necessary for bandit_feedback_test as it might be helpful in some cases. It should be int(self.n_rounds - n_rounds_train)

• sample_bootstrap_bandit_feedback( should be modified to adjust to the change
  https://github.com/nmasahiro/zr-obp/blob/b0700f0857dddfb6b81be1edb5adfa2db58b7bd9/obp/dataset/real.py#L263-L307

I will modify the benchmark scripts later by myself.

[nomuramasahir0]

Thank you for the review.

• I fixed the above points.
• In addition to this, I remove reward_test from the files.

Could you check it, please?

[usaito]

@nmasahiro Thanks! LGTM!