[fix]Avoid Floating point exception (core dumped) when using RedisTable by empty ids inputs.

[MoFHeka]

Description

Avoid Floating point exception (core dumped) when using RedisTable by empty ids inputs. I'm ignoring that and that's what's going to happen dividing by zero.

Also make compatible with the optimizers inited from tf.keras.optimizers. Optimizers call from tf.keras.optimizers are not belong to the class optimizer_v2.OptimizerV2 when using tf2.6.

Add TF2.7.0 CI.
Add CUDA_COMPUTE_CAPABILITIES.
Delete Python 3.6 in CI.

Make compatible with TF2.7 in reduction_ops_impl.h
Tensorflow2.7 requires C++14 or above, changing the standard for CUDA compilation from C++11 to C++14.

Type of change

• Bug fix
• New Tutorial
• Updated or additional documentation
• Additional Testing
• New Feature

Checklist:

• I've properly formatted my code according to the guidelines
  • By running yapf
  • By running clang-format
• This PR addresses an already submitted issue for TensorFlow Recommenders-Addons
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works

How Has This Been Tested?

About RedisTable

import numpy as np
import tensorflow as tf
import tensorflow_recommenders_addons as tfra

redis_config = tfra.dynamic_embedding.RedisTableConfig(
    redis_config_abs_dir="test/config1.json")
redis_creator = tfra.dynamic_embedding.RedisTableCreator(redis_config)

var = tfra.dynamic_embedding.get_variable(
        name="user_dynamic_embeddings",
        dim=32,
        initializer=tf.keras.initializers.Zeros(),
        checkpoint=False,
        kv_creator=redis_creator)

embeddings, trainable_wrapper = tfra.dynamic_embedding.embedding_lookup(
        params=var,
        ids=np.array([],dtype=int),
        name="user-id-weights",
        return_trainable=True)

print(embeddings)

The printing output will be looked like this

tf.Tensor([], shape=(0, 32), dtype=float32)

About optimizer
When using optimizer like this would not raise a error.

optimizer = tf.keras.optimizers.Adam(learning_rate=1E-4, amsgrad=True)
optimizer = tfra.dynamic_embedding.DynamicEmbeddingOptimizer(optimizer)

[Lifann]

I proposed an issue: tensorflow/tensorflow#53686

[rhdong]

This line is too long , make it multi-lines.