fix(abi): preserve array alignment and apply 16-byte workaround generically

[Snehal-Reddy]

This fixes an IllegalAddress error when passing 16-byte aligned arrays (e.g. [u128; 4]) or tuples as kernel parameters.

Previously, arrays indiscriminately forced PassMode::Direct(ArgAttributes::new()), discarding any alignment metadata computed by rustc. Furthermore, the 16-byte PassMode::Cast workaround was only applied to ADTs.

This change abstracts the 16-byte workaround to apply to any aggregate type (arrays, ADTs, tuples) with >=16-byte alignment, and properly preserves alignment metadata for arrays when PassMode::Direct is used.

Summary

This PR fixes PTX ABI code generation for arrays and tuples that require 16-byte (or greater) alignment. It resolves an internal compiler error (entered unreachable code: Align is given as power of 2 no larger than 16 bytes) and runtime IllegalAddress traps when launching kernels that take types like [AlignedStruct; 2] or large tuples.

Closes #387

Changes

• Introduced an is_aggregate condition that encompasses ADTs, Arrays, and Tuples.
• Applied the existing 16-byte PassMode::Cast workaround to all aggregate types.
• Updated the fallback PassMode::Direct assignment to correctly inherit arg.layout.align.abi, preventing alignment metadata from being blindly discarded.

Testing

• cargo build passes
• cargo clippy --workspace passes
• Tested on: Ubuntu 24.04.1 LTS, NVIDIA GeForce RTX 5060 Ti, CUDA 12.8.93

[Snehal-Reddy]

#389 should hopefully fix the CI?

[Snehal-Reddy]

Closed #389, #386 should now take care of windows CI failures.