add warning to adascale before it is validated (#169)

docs/source/tutorials/adascale.rst

 AdaScale SGD
 ============
 
+Note, AdaScale is still experimental. It is being validated. APIs may change
+in the future. Use at your own risk.
+
 `AdaScale <https://arxiv.org/pdf/2007.05105.pdf>`_ adaptively scales the learning rate when using larger batch sizes for data-parallel training. Let's suppose that your trainer looks like
 
 .. code-block:: python

fairscale/optim/adascale.py

@@ -32,6 +32,7 @@
 # POSSIBILITY OF SUCH DAMAGE.
 
 import functools
+import logging
 from typing import Any, Dict, Optional
 
 import numpy as np
@@ -79,6 +80,8 @@ class AdaScale(object):
         smoothing: float = 0.999,
         patch_optimizer: bool = False,
     ):
+        logging.warn("AdaScale is experimental. APIs may change. Use at your own risk.")
+
         self._optimizer = optimizer
         self._optimizer_step = optimizer.step
         self._local_grad_sqr: Optional[torch.Tensor] = None