[PyTorch] Expose `multi_tensor_*` kernels (#907)

* expose multi_tensor_* kernels
Signed-off-by: Xin Yao <xiny@nvidia.com>

* fix lint
Signed-off-by: Xin Yao <xiny@nvidia.com>

---------
Signed-off-by: Xin Yao <xiny@nvidia.com>

transformer_engine/pytorch/optimizers/__init__.py

@@ -3,6 +3,15 @@
 # See LICENSE for license information.
 
 """Fused optimizers and multi-tensor kernels."""
+from transformer_engine_torch import (
+    multi_tensor_scale,
+    multi_tensor_l2norm,
+    multi_tensor_unscale_l2norm,
+    multi_tensor_adam,
+    multi_tensor_adam_capturable,
+    multi_tensor_adam_capturable_master,
+    multi_tensor_sgd,
+)
 from .fused_adam import FusedAdam
 from .fused_sgd import FusedSGD
 from .multi_tensor_apply import MultiTensorApply, multi_tensor_applier