Fix dot-general contraction axis order

[shinaoka]

Summary

• keep GEMM dimension-group fusion in canonical col-major logical order
• add a dot_general regression for full contractions where RHS contracting axes are reversed

Verification

• cargo test -p strided-einsum2 dot_general_full_contract_respects_rhs_contracting_order -- --nocapture
• cargo fmt --check
• cargo test -p strided-einsum2 --features faer -- --nocapture
• cargo test -p strided-kernel --features parallel -- --nocapture
• cargo llvm-cov --workspace --json --output-path coverage.json && python3 scripts/check-coverage.py coverage.json
• git diff --check