fix(mir-lower): preserve address spaces through 16-byte enum fillers

[YunusMutlu]

Summary

Preserve known global pointer address spaces when lowering niche-encoded enums with a layout-neutral 16-byte filler. Representing the filler as [2 x i64] instead of [16 x i8] allows aggregate scalarization to retain the pointer's global address space, restoring ld.global / st.global codegen for the affected vector load and store paths.

Fixes #1331

Changes

• Represent layout-neutral 16-byte enum fillers as [2 x i64].
• Keep the byte-array fallback when the required 8-byte alignment is not available.
• Add enum layout coverage for 8-byte and 16-byte alignment cases.
• Add vectorization regressions for both load and store paths through a wide Option payload.
• Verify that the affected operations remain ld.global.v4.b32 and st.global.v4.b32.

Testing

• just check passes (the local mirror of CI: fmt, clippy, tests, guards, docs)
• cargo oxide run vectorization --arch sm_86 passes
• New example added (if applicable)

Additional validation:

• cargo test -p mir-lower
• cargo clippy --workspace --all-targets -- -D warnings
• (cd crates/rustc-codegen-cuda && cargo clippy --all-targets -- -D warnings)
• git diff --check
• Verified on sm_86 that the load changes from ld.v4.b32 / LD.E.128 to ld.global.v4.b32 / LDG.E.128
• Verified on sm_86 that the store changes from st.v4.b32 / ST.E.128 to st.global.v4.b32 / STG.E.128

Checklist

• All commits signed off (git commit -s)
• SPDX headers on new source files

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.