[quant][gpu][core][bug fix] Added memset to CacheKey for quantized cudnn linear op (#76447)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/76447
See https://github.com/pytorch/pytorch/pull/76436. The same idea is
applied here.
Test Plan:
```
python test/test_quantization.py -k test_qlinear_cudnn
```
Reviewed By: jerryzh168
Differential Revision: D35970360
Pulled By: dzdang
fbshipit-source-id: 452f0703d3c257638d2ef1e2467756aeed0ec8d4
(cherry picked from commit ff52d26ad477bd2c0529b627ea3cc763523f80e6)
