Update to XGBoost 1.0.2

[jeremyliweishih]

Currently XGBoost 1.0.0 breaks our python 3.5 tests and make the others timeout. We should at least cap XGBoost for now and have further discussion on how we handle dependency versioning.

[dsherry]

@jeremyliweishih should we attach #407 to this ticket?

[dsherry]

@christopherbunn is gonna look into this a bit more this afternoon, but after that he's out until next week. So I'll pick it up tomorrow.

First off, we no longer need to exclude xgboost 1.0.0 because we no longer support python 3.5. Updating issue title to reflect that.

Regarding the errors: I just spoke with @christopherbunn and we see the following when running xgboost 1.0.1:

• Running xgboost pipeline during automl (for example test_random_state) takes ~9min to complete. That's causing that test to time out, because that test runs automl 4x and there's a 20min timeout
• @christopherbunn observed xgboost was using 26 cores when configured with n_cores=-1 instead of 8. We think 26 is the number of cores on the circleci machine
• We're still not sure if this is related to the errors @angela97lin is seeing in [WIP] Allow numpy.random.RandomState for random_state #441, or to the failure in Flaky unit tests: joblib parallel ValueError #167. Although it is clear all these issues have to do with n_jobs/joblib.