Implement native flat array algorithm

Add fully-flat data path that avoids 2D<->1D conversion overhead:
- generate_random_flat / generate_glider_flat
- next_generation_flat (takes/returns arrayref)
- print_grid_braille_flat
- count_live_cells_flat

Results on 100x100 grid, 1000 generations (no display):
- scalar:   214 gen/s (baseline)
- parallel: 1104 gen/s (5.2x)
- flat:     1132 gen/s (5.3x, +2.5% vs parallel)

The flat array provides a small speedup for computation-only
workloads by avoiding array-of-arrays dereference overhead.

Generated with [Devin](https://cli.devin.ai/docs)

Co-Authored-By: Devin <noreply@cognition.ai>

Author: fglock