fix(ci): satisfy Rust 1.98 byte-chunk lint

[GiggleLiu]

Observation

cargo xtask check lint fails on Rust 1.98 in read_host_data for both F32 and I32 data:

error: use of `chunks_exact` with `try_into`
help: consider using `as_chunks`

CubeK runs Clippy with --deny warnings, so this warning blocks CI.

Root cause

Both decoding loops split a byte buffer into four-byte slices with chunks_exact(4), then convert each slice to [u8; 4] with try_into().unwrap().

The chunk size is known at compile time, and each buffer is allocated as elem_count * size_of::<T>() before read_exact fills it. There cannot be a remainder. Rust 1.98 Clippy therefore flags the fallible slice conversion and asks for as_chunks::<4>().

Fix

Use buf.as_chunks::<4>().0 and pass each [u8; 4] directly to from_le_bytes.

This preserves the decoding behavior, removes the unnecessary conversion and unwrap, and leaves the CubeCL dependency pin unchanged.

Validation

• cargo fmt --all -- --check
• cargo test -p t4a-cubek-test-utils --lib round_trip (3 passed)
• cargo xtask check lint