[dnnlowp] fix qparam handling during bias quantization (#73478)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/73478
temp_qparams is not used and while constructing it accessing non-existent qparams
Test Plan: buck test mode/dev -c fbcode.platform=platform010 //caffe2/caffe2/quantization/server:conv_groupwise_dnnlowp_acc16_op_test -- --exact 'caffe2/caffe2/quantization/server:conv_groupwise_dnnlowp_acc16_op_test - test_groupwise_dnnlowp_conv_acc16_outlier (caffe2.caffe2.quantization.server.conv_groupwise_dnnlowp_acc16_op_test.GroupWiseDNNLowPOpConvAcc16OpTest)'
Reviewed By: jiyuanzFB
Differential Revision: D34502856
fbshipit-source-id: aeaa5c3aa76a2fb01d9565ee294a0c627418f55e
(cherry picked from commit 9f930156b3823d7f0676e26ebd4d0ae1e682f08b)
