Add instance norm

[zsef123]

Description

Add InstanceNorm usingTensorRT InstanceNormalizationPlugin and implementation of BatchNorm

Fixes #210

Also I checked #285
But this PR not supported TensorRT 8++
And it's a little different from mine.

1. default beta value, BN and feat(//core/): Add support for tensorrt plugin and implement instance… #285 set to 1 but I think that must be 0
2. type of eps, float is right type. And also my env, double occurs error.
3. I called TensorRT directly not modifying other context (follows gelu)

Type of change

Please delete options that are not relevant and/or add your own.

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

Checklist:

• My code follows the style guidelines of this project (You can use the linters)
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas and hacks
• I have made corresponding changes to the documentation
• I have added tests to verify my fix or my feature
• New and existing unit tests pass locally with my changes

[zsef123]

@narendasan passed DCO

[narendasan]

@zsef123 Thanks for the contribution! Can you rebase your branch? Changes look mostly fine, once you rebase, I will run the tests and we should be good to go

[zsef123]

@zsef123 Thanks for the contribution! Can you rebase your branch? Changes look mostly fine, once you rebase, I will run the tests and we should be good to go

Checked!!

[narendasan]

I am getting test failures for Converters.ATenInstanceNormRunningStatsConvertsCorrectly where the TRT implementation returns NaN. Using TRT8

[zsef123]

@narendasan Sorry for missed.

Several problems were mixed. All of them were test cases.

1. range of running_mean and running_var
   In testcase, cause overflow when calculate gamma and bias.
   So changed randn -> ones, zeros. It is default value of instance_norm

2. not clone parameters in test case
   fixed

And changes

1. changed device
   It is not bug, but I just follows batch_norm
   Changed cpu to cuda when the track_running_stats==True

2. readable graph in test

P.S.
I had a hard time creating a new docker environment.
Any plans for a distribution with NGC 21.07?

[narendasan]

@zsef123 check #578 for a 21.07 container. It uses a version of Pytorch slightly newer than 1.9 but other than that should work

[narendasan]

Cool, tests pass, LGTM