Fix uninitialized record assignment in reverse mode

[Elvand-Lie]

Summary

• replace declaration-only empty constructor expressions with Clad’s existing zero initializer before building promoted assignments
• preserve explicit list initialization and constructor calls with arguments
• add regression coverage for record declarations inside loops

Testing

• verified the issue Gradient: Fix missing struct initialization to structs declared inside a loop but not originally initialized #1283 reproducer fails on origin/master and passes with this change
• manually A/B tested implicit aggregates, default member initializers, explicitly defaulted records, explicit list initialization, and user-defined constructors
• ran the full lit suite: 174 passed, 1 expected platform failure, 25 unsupported, and 2 failures independently reproduced on origin/master
• ran DifferentiatorHTests and MiscTests: 5/5 passed
• ran git-clang-format-18: no changes
• ran clang-tidy-18 on the changed production-code range: no diagnostics
• ran git diff --check

Fixes: #1283

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[vgvassilev]

Why not using CHECK: and CHECK-NEXT? We also need to check the execution results.

[Elvand-Lie]

Updated in a9fe9a5. The regression now uses CHECK/CHECK-NEXT for the generated loop, builds and runs the test executable, and checks the derivative result (6.00) with CHECK-EXEC. I also verified that the revised lit test passes with this patch and fails against current origin/master at the original s = ; line.

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"