[ShaderGraph] [2021.1/staging] [bugfix 1308156] Backport bugfix for view direction node in tangent space being not using the calculation

[joshua-davis]

Backport fix for #3044

Please read the Contributing guide before making a PR.

• Read the Graphics repository & Yamato FAQ.

Checklist for PR maker

• Have you added a backport label (if needed)? For example, the need-backport-* label. After you backport the PR, the label changes to backported-*.
• Have you updated the changelog? Each package has a CHANGELOG.md file.
• Have you updated or added the documentation for your PR? When you add a new feature, change a property name, or change the behavior of a feature, it's best practice to include related documentation changes in the same PR. If you do add documentation, make sure to add the relevant Graphics Docs team member as a reviewer of the PR. If you are not sure which person to add, see the Docs team contacts sheet.
• Have you added a graphic test for your PR (if needed)? When you add a new feature, or discover a bug that tests don't cover, please add a graphic test.

---

Purpose of this PR

The calculation for the ViewNode in Tangent space used an old method that assumed the tangent space basis was orthonormal. The Transform node was updated but the view node wasn't. For this backport, the length of the original view direction is preserved in case anyone was relying on that.

---

Testing status

Tested the attached shader graph in the bug: https://fogbugz.unity3d.com/f/cases/1296788/

---

Comments to reviewers

There looks to be a unrelated failed test with Android Vulkan due to an incorrect image in this branch (on a nan node).

[ghost]

Tested everything I tested in the original PR (except with normalization steps added). Tested that the length of the tangent space view dir node matched the world space length.
Results of additional test to validate that the length matches the original length "close enough":

After on the left, before on the right, with the length of the vector mapped the hue in an HSV to RGB colorspace conversion. Users can expect extremely similar results, but with less artifacts caused by the previously incorrect calculation.

[jessebarker]

LGTM

[joshua-davis]

Failures seem to match the current state of upstream (2021.1/staging)