Add Gaussian NLL Loss

[nailimixaM]

Fixes #48520.

cc @albanD (This is a clean retry PR #49807)

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit 2204266 (more details on the Dr. CI page):

---

• 2/2 failures possibly* introduced in this PR
  • 2/2 non-CircleCI failure(s)

---

ci.pytorch.org: 2 failed

• Failed: pr/caffe2-pytorch-linux-bionic-rocm3.10-py3.6-test
• Failed: pr/pytorch-linux-bionic-rocm3.10-py3.6

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

[albanD]

Looks great!
Could you just revert the tensorpipe submodule update please? (let me know if you want me to do it)

Will land when the submodule is fixed!

[nailimixaM]

Looks great!
Could you just revert the tensorpipe submodule update please? (let me know if you want me to do it)

Will land when the submodule is fixed!

Done, fingers crossed this works!

[albanD]

(rocm failures look unrelated)

[facebook-github-bot]

@albanD has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[codecov]

Codecov Report

Merging #50886 (2204266) into master (480bb7d) will increase coverage by 0.00%.
The diff coverage is 84.21%.

@@           Coverage Diff           @@
##           master   #50886   +/-   ##
=======================================
  Coverage   81.01%   81.01%           
=======================================
  Files        1916     1916           
  Lines      209285   209323   +38     
=======================================
+ Hits       169562   169593   +31     
- Misses      39723    39730    +7     

[facebook-github-bot]

@albanD merged this pull request in 8eb90d4.

[almson]

Please see #53964

Main points:

• reduction 'none' is not implemented the way it is elsewhere
• nn.functional.gaussian_nll_loss is missing in the documentation
• full parameter is useless and not theoretically grounded.
• var should not be clamped in a no_grad context. gradients should be masked by the clamping.

[nailimixaM]

@almson Thanks for your points. I've addressed these in #53964.