fix(warp): properly interpret floating point numbers in match_any

[Snehal-Reddy]

Summary

This PR fixes a critical logic bug in WarpMatchValue where f32 and f64 types were being incorrectly processed by the generic impl_match! macro.

The macro dynamically applies integer casting (value as u32 and value as u64) to cast inputs to bitmasks before executing the underlying SIMT PTX intrinsic. While this acts as a safe bitwise transmute for values like i32, in Rust, casting a float via as u32 executes a truncating numeric conversion. For example, both 1.1f32 and 1.9f32 truncate to 1u32, and all negative floats cast identically to 0. Consequently, operations like warp_match_any between completely distinct floats would silently evaluate as a perfect match, catastrophically corrupting thread convergence checks and logic algorithms expecting structural equality.
Closes #406

Changes

• Removed the f32 and f64 types from the impl_match! macro in crates/cuda_std/src/warp.rs to prevent integer cascading behavior.
• Implemented WarpMatchValue manually for f32, strictly leveraging value.to_bits() to route structural memory bits into match_any_32 and match_all_32.
• Implemented WarpMatchValue manually for f64, strictly leveraging value.to_bits() to route structural memory bits into match_any_64 and match_all_64.

Testing

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