Remote filesystem support, estimator fixes

PyTorch Lightning support, Nsight profiling, PyTorch 1.9 compatibility

Hotfix for TensorFlow 2.4, grouped allreduce updates

PyTorch 1.8 compatibility, Spark on Gloo stdout forwarding, bugfixes

Bugfixes for Ray, Databricks, and Spark Estimators

Local gradient aggregation, grouped allreduce

Elastic Horovod on Ray

Hotfix: build without MXNet installed

Bugfixes, Databricks support for Estimators

Elastic Horovod, Ray integration, CMake build, all-to-all op, gradien…

…t predivide factor