GroupNorm + FP16 instability

[shkarupa-alex]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Colab
• TensorFlow version and how it was installed (source or binary): 2.6.0
• TensorFlow-Addons version and how it was installed (source or binary): 0.14.0
• Python version: 3
• Is GPU used? (yes/no): no

Describe the bug In some cases GroupNormalization layer generates Inf/Nan values with mixed_fp16 policy.

Code to reproduce the issue https://colab.research.google.com/drive/17zqRy30dbOcJpWZfOwZK3UFovIJZCuqb?usp=sharing

[bhack]

/cc @smokrow