Probability values still incorrect when requesting raw and probability columns

[imatiach-msft]

Describe the bug
There is a caching bug in the new native code which causes the probability values to be calculated as raw when requesting both columns to be computed.

To Reproduce
Print out the results of lightgbm model, see that the probabilities are invalid, eg not between 0 and 1

Expected behavior
Probabilities should correspond to output label and raw predictions transformed from log odds.

Info (please complete the following information):

• MMLSpark Version: v0.18.1
• Spark Version Any
• Spark Platform Any

Please post the stacktrace here if applicable

If the bug pertains to a specific feature please tag the appropriate CODEOWNER for better visibility

Additional context
See lightgbm PR here:
microsoft/LightGBM#2356
a new maven dependency will need to be published once it is in.

See related issue, but fixed in correct way as long as user only asks for one column:
#568

[imatiach-msft]

fixed in master with #676