AppendCacheCheckpoint is ignored at the beginning of the pipeline

[yaeldekel]

When an empty EstimatorChain is created, and AppendCacheCheckpoint is called before adding any estimators, then no caching occurs.
The expected behavior should be to either cache the data before fitting the first estimator in the chain, or disallow calling AppendCacheCheckpoint on empty chains, so that the user knows to cache the data explicitly before fitting. (I think the first solution is preferable, but open to other opinions).

[artemiusgreat]

And this is the code to reproduce the issue :)
#4926 (comment)

[frank-dong-ms-zz]

This issue is different issue with issue #4926, issue #4926 is converting from EstimatorChain to IEstimator and lose information like need cache. This issue is when try to call AppendCacheCheckpoint after an empty EstimatorChain and no cache will happen.