Fix Conv2dTranspose bias

test: Gradient descent  optimizer tests

passing

fix: fix the validation_pack when multiple input

refactor: refacter the place of the kernel unittest folder

fix: fix the bug caused by concat_v2

test: add the concat_v2 test

fix: fix the bug of boolean_mask

fix: add the implementation of the tile's and GatherND's grad and add OptionalArgs

fix: Saving a loaded model

add relu6

fix: Implemented support for loading models with Concatenate layers

The loading and saving of a simple model with a Concatenate layer is tested to check if the model is the same after reloading.
Implemented missing axis parameter for np.stack (added some handy tuple calls too like the np.concatenate example).

model.load_model now supports loading of concatenate layers.
python tensorflow exports concatenate layers in an extra nested array in the manifest so added a check for that in generic_utils.cs.
Concatenate was missing the build=true, this fix prevents the layer being build multiple times.
Concatenate has 2 or more input nodes so List<NodeConfig> was required instead of just NodeConfig in Functional.FromConfig.cs.
Added missing axis JsonProperty attribute for MergeArgs (used by Concatenate)

解决keras模式下，使用GPU训练时会爆显存的bug。

观察到的现象是，一些模型增大batchsize后，会在首个epoch的中途爆显存不足，只要过了一个epoch后，就能完整训练。同样的batchsize在python下能设置大得多的值。
最后使用最小训练代码分析出，是每个step之后，图片加载到显存里的数据没有释放导致的。
在寻找释放显存接口没有结果的时候，直接使用了GC.Collect();可以让显存主动回收。
因此当前的修复方案是在每个step里，都执行一次 GC.Collect(); 用来释放显存资源。

feat: add the implementation of class_weight in model.fit

Allow Model to cache weights.

feat: add the implementation of sample_weight in model.fit