[PR #1537 Patch 4] Separate users' evaluation metric and the internal validation metric

[zhiqiangdon]

This PR is to separate users' provided evaluation metric and the internal validation metric. If a user provides metric r2, we will use rmse as the internal validation metric since torchmetrics.R2Score encounters errors for batch size 1.

1. Always keep the evaluation metric provided by users. It can be used in .evaluate().
2. Use a separate internal validation metric, which may be different from the provided evaluation metric.
3. Rename the names of evaluation and validation metrics so that their names are both distinguishable and accurate, especially in the context of deep learning.
4. Clean some unused log keys.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[szha]

Job PR-1571-2 is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-1571/2/index.html

[szha]

Job PR-1571-3 is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-1571/3/index.html