ci: add FFT and MPI benchmark coverage

[ulises-jeremias]

benchmarks/ covers BLAS, LAPACK, and NumPy only. FFT and MPI have no benchmarks.

Task

1. Add benchmarks/fft_bench.v comparing vsl.fft against numpy.fft
2. Add benchmarks/mpi_bench.v for MPI send/reduce/broadcast latency
3. Wire both into benchmarks.yml matrix

Files

• New benchmarks/fft_bench.v
• New benchmarks/mpi_bench.v
• .github/workflows/benchmarks.yml