Minor quality-of-life and performance enhancements

[asuradev99]

• Added global_swap MCMC mode for vacancy-free swaps between randomly selected atoms of different chemical species.
• Made global_swap bypass vacancy detection, lattice construction, and nearest-neighbor searches (standard MCMC Metropolis hastings atom swapping algorithm)
• Preserved the existing behavior of vacancy_only, fix_vacancy, all, LSS, and TKS modes.
• Reused a single ASE/MACE calculator instead of reconstructing it when atomic configurations change (unnecessary).
• Cached the current accepted-state energy during MCMC, removing redundant energy evaluations.
• Added max_iter = 0 support for disabling structural relaxation.
• Added buffered XDATCAR writing with configurable trajectory_buffer_size; the default is 100 frames.
• Added optional MACE cuEquivariance acceleration through enable_cueq; testing on my MCMC config gave ~2x speedup.
• Added validation preventing output_name from being empty or containing a path.
• Added configs/deploy_mcmc_example.toml documenting the new MCMC options.
• Added only the cuEquivariance dependency versions to the README