LSTM model init failure

[ziggy1209]

1. 环境（environment）

• Build OS and Version: Ubuntu 20.04 focal
• RunTime OS Version: Linux

• RunTime DEVICE: ARM

  2. Github版本

• branch：Not quite sure about this since I directly pulled the docker image using docker pull ccr.ccs.tencentyun.com/qcloud/tnn-convert

• commit(optional):

  3. 编译方式(compile method) CMake完整编译参数(full cmake arguments) Not applicable as I used docker image

  4. 编译日志(build log)

Not applicable

5. 详细描述bug 情况 (Describe the bug) Converted a LSTM model successfully but encountered layer init failure during inference session

6. 运行日志（runtime log）

Initializing Model...
E/tnn: RawBuffer2ArmBlob [File source/tnn/device/arm/acc/arm_layer_acc.cc][Line 135] RawBuffer2ArmBlob:: unsupported buffer and blob data type: 3, 0
E/tnn: InitLayers [File source/tnn/core/default_network.cc][Line 333] Error Init layer Expand_30 (err: 4096 or 0x1000)

7. 截图（Screenshots）

![Screenshot from 2023-03-31 18-10-40](https://user-images.githubusercontent.com/58074506/229092057-a7e08f0f-8ff1-481a-ab35-959ce76cf78f.png)

[ziggy1209]

Found the problem to be expanding a 1D vector to 3D which is not supported by TNN ARM.