[MNN:Sync] A few bugfixes

    1. 支持 Onnx If 空子图的情况（这种情况是条件判断一定为真或假）
    2. 修正 Where 算子在 zeroshape 下维度计算出错的问题
    3. 修正 Reduce 计算 zeroshape 的非 prod 情况
    4. 修正 arch64-linux 上编译错误
    5. 修正 头文件 NNAPI 的注释错误
    6, 部分训练相关问题修正

[Sync] Sync Internal 2.2.2

[Sync] Sync Internal Gitlab 2.2.1 contains below commits:

[Sync] Sync Internal Gitlab 2.2.0

[Sync] Sync Internal 2.1.2

Add Interp3D Support

merge interp3d_op_param into interp_op_param

Add ConvTranspose3D

accelerate MNNPackTranspose

fix bugs for leakrelu in hiai ir for mnn ir

maintain for hiai backend

fix compiling error for hiai

MNN_BUILD_SHARED_LIBS=OFF build OnnxClip

Fix MNN_BUILD_PROTOBUFFER=OFF

Fix ThreadPool behavior

converter caffe adapt Eltwise

feat(iOS): add Apple A16 support in CPU family

[MNN:Sync] Sync Internal 2.1.1 contain below changes.

    [Pymnn:Bugfix] Fix usage and small bug in pymnn.
    [Docs:Update] Update docs/cpp markdown
    [Docs:Update] Add docs check.
    [MNN:Update] Update VecHalf.hpp
    [MNN:Bugfix] Fix compile errors caused by "#define MNN_THREAD_LOCK_CPU"
    [Geometry:Bugfix] Fix bug for resize of broadcastto: https://github.com/alibaba/MNN/issues/2040
    [Docs:Update] Update inference api usage.
    [Pymnn:Bugfix] Close hiai load to fix resource leak.
    [MNN:Update] Down gradle version for demo compile

In ONNX, when neural network is Conv3D -> Clip(1, -1), the converter will convert these two OPs in to Conv3DRelu6, regardless the min max values are not 0 and 6. This patch fix this bug.

 Pipeline::encode -> GeometryComputerUtils::shapeComputeAndGeometryTransform -> onExecute

1. add PyTorch interpolation 3D to Onnx to MNN converter
2. add interpolation3D nearest CPU/OpenCL implementation

all added OPs are verified
update opencl_program.cc

update opencl_program.cc