[SYCLomatic]Remove reverse_iterator from partition implementations

[danhoeflinger]

Partition implementations should not use std::reverse_iterator, because this disallows usage of sycl buffer iterator-like inputs returned from oneapi::dpl::[begin,end]. MSVC STL runs into compiler errors for reverse_iterator of this iterator-like type.

Without a reverse iterator, we no longer can gain advantage from providing a non-stable partition implementation, so we go back to having a unified implementation (these were separated in #2215). If in the future oneDPL provides a reverse_iterator which is compatible with sycl buffer iterator-like types, we can improve this implementation by removing this extra copy call, and extra tmp data allocation. However, for now I think this is the best we can do.

[timmiesmith]

LGTM. Thank you.

[mmichel11]

LGTM. We may want to create an issue somewhere about this problem if we want to consider adding a reverse iterator to oneDPL in the future.

[danhoeflinger]

LGTM. We may want to create an issue somewhere about this problem if we want to consider adding a reverse iterator to oneDPL in the future.

uxlfoundation/oneDPL#1836

[tomflinda]

LGTM