chore(release): bump t4a CubeCL packages to 0.10.1

[shinaoka]

Why

The published t4a-cubecl* 0.10.0 crates predate fix(cpp): lower CUDA complex casts with cuComplex helpers (#17). Assign::format_scalar in the published t4a-cubecl-cpp 0.10.0 still emits a plain C++ constructor call for every scalar cast:

match elem {
    Elem::TF32 => write!(f, "nvcuda::wmma::__float_to_tf32({input})"),
    elem => write!(f, "{elem}({input})"),
}

so any kernel that casts a real value into a complex dtype fails NVRTC compilation with

error: no suitable constructor exists to convert from "uint32" to "double2"
  const cuDoubleComplex l_16 = cuDoubleComplex(uint32(0));

This is what tensor4all/tenferro-rs#1833 reports: complex triu/tril, and therefore complex QR, are unavailable on CUDA for every crates.io consumer of tenferro-rs 0.5.0. Repository builds do not see it because tenferro-rs pins this fork by git rev (a2adda17), which already contains the fix.

What

Bump [workspace.package].version and every first-party inter-crate requirement from 0.10.0 to 0.10.1 so the fix can be published. Third-party spin and rand requirements that also read 0.10.0 are left untouched.

After this merges, tagging v0.10.1 publishes the crate set listed in AGENTS.md through .github/workflows/publish.yml, and tenferro-rs can move its pin to =0.10.1.

Verification

• cargo metadata --no-deps resolves all 13 first-party packages at 0.10.1
• no source changes; the CUDA complex-cast lowering already on main is what this release carries

🤖 Generated with Claude Code