Depth clip/clamp test

[kainino0x]

See gpuweb/gpuweb#2100

In Chromium (except format we don't support):

• Passing on Metal
• Passing on D3D12
• Failing on Vulkan because we need to add a code transform: https://crbug.com/dawn/1125

---

Author checklist for test code/plans:

• All outstanding work is tracked with "TODO" in a test/file description or .unimplemented() on a test.
• New helpers, if any, are documented using /** doc comments */ and can be found via helper_index.txt.
• (Optional, sometimes not possible.) Tests pass (or partially pass without unexpected issues) in an implementation. (Add any extra details above.)

Reviewer sign-off for test code/plans: (Note: feel free to pull in other reviewers at any time for any reason.)

• The test path is reasonable, the description "describes the test, succinctly, but in enough detail that a reader can read only the test plans in a file or directory and evaluate the completeness of the test coverage."
• Tests appear to cover this area completely, except for outstanding TODOs. Validation tests use control cases.
  (This is critical for coverage. Assume anything without a TODO will be forgotten about forever.)
• Existing (or new) test helpers are used where they would reduce complexity.
• TypeScript code is readable and understandable (is unobtrusive, has reasonable type-safety/verbosity/dynamicity).

[github-actions]

Previews, as seen when this build job started (1d6a59c):
Run tests | View tsdoc

[github-actions]

Previews, as seen when this build job started (bf076c7):
Run tests | View tsdoc

[kainino0x]

PTAL! @kvark @Jiawei-Shao since you've been following gpuweb/gpuweb#2100, perhaps you could both review these (rather nitpicky) tests? Please see if there's aspect I've missed testing.

@Jiawei-Shao, perhaps additionally you could verify these tests against an implementation of the depth-clamping feature on D3D12, since I think you were working on that?

These tests are passing in Chromium (except for depth16unorm which we don't implement yet) on Metal (with both depthClamp values) and D3D12 (with depthClamp:false). They pass on Vulkan with clampDepth:true but fail with clampDepth:false, because for consistency with D3D12/Metal, Tint will need to inject a clamp to the viewport bounds at the end of fragment stages that write frag_depth.
I haven't tried them against Firefox because I don't know what the implementation status is.

These tests will need updating later because I think we need to change the spec from having a depthClamp toggle to having a depthClip toggle.

[kainino0x]

Also I tried to write good comments, but please let me know if any of it can be clarified further.

[kainino0x]

Thanks!