Compat issue with Taichi - cuda_mapped_frame.cc: invalid resource handle

Similar to #227, when I run this code

```python
import taichi as ti
import decord as d

ti.init(arch=ti.cuda)

mp4 = d.VideoLoader(["../../data_out/9fd43132-8c77-4c4f-a2e0-725bfde186d1.mp4"], ctx=d.gpu(), shape=(32, 1080, 1920, 3), interval=1, skip=0, shuffle=0)

next(mp4)
```

I get the error message

```
[Taichi] version 1.1.2, llvm 10.0.0, commit f25cf4a2, win, python 3.10.5
[Taichi] Starting on arch=cuda
[W 09/15/22 09:57:03.284 14716] [D:/a/taichi/taichi/taichi/runtime/llvm/llvm_offline_cache.cpp:load_meta_data@98] Lock C:/taichi_cache/ticache/metadata.lock failed
[09:57:03] C:\Users\rb\Development\Source\_thirdparty\decord\src\video\nvcodec\cuda_threaded_decoder.cc:36: Using device: NVIDIA GeForce RTX 2080 Ti
[09:57:03] C:\Users\rb\Development\Source\_thirdparty\decord\src\video\nvcodec\cuda_threaded_decoder.cc:56: Kernel module version 516.94, so using our own stream.
CUDA error 400 at line 35 in file C:\Users\rb\Development\Source\_thirdparty\decord\src\video\nvcodec\cuda_mapped_frame.cc: invalid resource handle
```

Setup is still

- Windows 11
- Python 3.10 (Mambaforge)
- Decord d2e56190286ae394032a8141885f76d5372bd44b compiled with CUDA support
- Taichi 1.1.2