Revert "feat: support DuplicatedNoNeed as function annotation for Enzyme"

[gdalle]

Reverts #805

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.93%. Comparing base (70e91d7) to head (eaae10a).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #807      +/-   ##
==========================================
- Coverage   97.93%   97.93%   -0.01%     
==========================================
  Files         128      128              
  Lines        7697     7693       -4     
==========================================
- Hits         7538     7534       -4     
  Misses        159      159              

Flag	Coverage Δ
DI	98.81% <100.00%> (-0.01%)	⬇️
DIT	95.61% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[wsmoses]

why was this reverted?

[gdalle]

Because the tests passed but I don't understand why. Despite the follow-up exchange on Discourse I still don't fully get what I'm allowed to do with DuplicatedNoNeed, nor can I think of a meaningful test case where the closure storage is write-only (what good does a write-only storage do?).
The test case I had in the DI test suite writes to closure storage and then reads back from it, but switching to function_annotation=DuplicatedNoNeed did not make it fail. I'm not sure why, so as a precaution I'm putting correctness over performance until I figure it out.

[gdalle]

Essentially I don't want to release something where tests are supposed to fail but succeed and I don't know why.

[gdalle]

The test case in question is here:

https://github.com/JuliaDiff/DifferentiationInterface.jl/blob/main/DifferentiationInterfaceTest/src/scenarios/modify.jl#L108-L164

I take a standard test function, and then I force both the input and the output to go through a buffer