[codex] Fix ComplexF64 SimpleTT site tensor regression

[shinaoka]

Summary

This PR supersedes the conflict-heavy part of #22 with a clean branch from current main.

Changes:

• fix t4a_simplett_c64_site_tensor buffer-length handling in the Julia C wrapper so it passes the number of doubles expected by the Rust C API
• add test/test_conversions.jl coverage for SimpleTensorTrain <-> TreeTN round-trips, including a ComplexF64 regression that exercises the previously broken path

Root cause

Tensor4all.jl passed the complex site-tensor buffer length as the number of complex elements, but the Rust C API expects the number of interleaved Float64 values. That caused SimpleTT complex site-tensor extraction to fail with T4A_INVALID_ARGUMENT, which in turn broke the ComplexF64 SimpleTT and TreeTN conversion path.

Validation

Passed:

• julia --startup-file=no --project=/home/shinaoka/tensor4all/Tensor4all.jl -e 'using Test, Tensor4all, Tensor4all.SimpleTT; include("/home/shinaoka/tensor4all/Tensor4all.jl/test/test_simplett.jl")'
• julia --startup-file=no --project=/home/shinaoka/tensor4all/Tensor4all.jl -e 'using Test, Tensor4all, Tensor4all.SimpleTT, Tensor4all.TreeTN; include("/home/shinaoka/tensor4all/Tensor4all.jl/test/test_conversions.jl")'
• T4A_SKIP_HDF5_TESTS=1 julia --startup-file=no --project=/home/shinaoka/tensor4all/Tensor4all.jl /home/shinaoka/tensor4all/Tensor4all.jl/test/runtests.jl

Result:

• 293 passed
• 0 failed
• 0 broken

Related

• supersedes the relevant fix portion of feat: API expansion and quantics Rust parity #22
• depends on the current Rust C API behavior already present on main