[MRG] Update the advisor to support various modes

[huangyz0918]

precise mode is more suitable for a task with clear metrics and datasets.

Here is an example of using auto_kaggle to complete a Kaggle competition:

mle kaggle --auto \
--datasets "/Users/huangyz0918/Library/Caches/mle-bench/data/spooky-author-identification/prepared/public/train.csv,/Users/huangyz0918/Library/Caches/mle-bench/data/spooky-author-identification/prepared/public/test.csv" \
--description "/Users/huangyz0918/Library/Caches/mle-bench/data/spooky-author-identification/prepared/public/description.md" \
--submission "/Users/huangyz0918/desktop/test-kaggle3/submission.csv" \
--sub_example "/Users/huangyz0918/Library/Caches/mle-bench/data/spooky-author-identification/prepared/public/sample_submission.csv" \
--comp_id "spooky-author-identification"

FYI @leeeizhang

[leeeizhang]

thanks! the precise mode is very important for kaggle workflow!