feat: 2D support

[nihiL7331]

This PR implements a macro-based 2D support, stripping out anything from the 3D implementation that isn't needed in 2D.

• Morton codes get 16 bit per dimension, so the range is far greater for 2D
• The memory usage is lower, two nodes fit in one cache line
• There's less recursive calls and traversal (octotree in 3D->quadtree in 2D)

Alongside the 2D implementation, fixed/refactored:

• multipole array got moved from ps_node_t to ps_context_t matching the local expansion structure,
• ps_impl_fmm_interaction_pass/ps_impl_fmm_p2p_pass subdivide logic is now size-based to ensure equal division,
• run_st_vs_mt_test got fixed to match the current approach in ps_calc_forces.

Also added:

• 2D tests, matching the main_3d.c tests (previously main.c),
• 2D unary demo, showcasing the 2D capabilities.

Demonstration code might need a rewrite. It's poorly written, and the renderer logic bottlenecks the physics logic. Rewriting the code in something like sokol with pushing the array of particles to the GPU might be the best bet.