MO should support LRN k param with caffe model, rather than fixed to 1

[wistal]

I found when mo convert caffe model, the k param of LRN layer was hardcoded to 1, which should read from prototxt.
for example:
layer {
name: "norm1"
type: "LRN"
bottom: "relu1"
top: "norm1"
lrn_param {
local_size: 5
k: 2.000000
alpha: 0.000500
beta: 0.750000
}
}
When convert to mo, k will be set to 1, correct value should keep to 2.
The issue seems only for mo caffe.

I fixed it by modify mo code lrn_ext.py, and verified using 2 modes, the conversion result is correct.

[yury-gorbachev]

@lazarevevgeny can you help?

[lazarevevgeny]

@yury-gorbachev, I have restarted the test pipeline since it failed due to timeout. Once it is passed I will merge the PR.

[lazarevevgeny]

@wistal, please, pick up changes from master.