Add fov_y parameter to equi2pers for accurate vertical-FOV intrinsic matrix

[Copilot]

Implements the TODO from the README:

More accurate intrinsic matrix formulation using vertical FOV for equi2pers

Also completes the documentation TODO for equi2pers.

What changed

New fov_y parameter

All equi2pers entry-points (Equi2Pers, equi2pers, get_bounding_fov) and the underlying create_intrinsic_matrix functions (numpy & torch) now accept an optional fov_y: float = None parameter.

fov_y	behaviour
None (default)	fy is derived from fov_x and the image aspect ratio — equivalent to the previous square-pixel assumption, fully backward-compatible
provided	fy = height / (2 * tan(fov_y / 2)) — lets callers specify vertical FOV directly or model non-square-pixel sensors

The intrinsic matrix now reads:

K = [[fx,  skew, cx],
     [0,   fy,   cy],
     [0,   0,    1 ]]

where fx = width / (2 * tan(fov_x / 2)) and fy is set independently when fov_y is given.

Files changed

• equilib/numpy_utils/intrinsic.py — add fov_y to create_intrinsic_matrix
• equilib/torch_utils/intrinsic.py — same, supports torch.Tensor for differentiability
• equilib/equi2pers/numpy.py — thread fov_y through create_cam2global_matrix, prep_matrices, run, get_bounding_fov
• equilib/equi2pers/torch.py — same; also includes fov_y in the mode='nearest' grad-warning check
• equilib/equi2pers/base.py — expose fov_y in Equi2Pers.__init__, equi2pers, and get_bounding_fov
• tests/test_torch_utils/test_grad.py — add test_intrinsic_matrix_fov_y_gradcheck and test_intrinsic_matrix_fov_y_grad_nonzero
• tests/test_equi2pers/test_base.py — add test_numpy_fov_y and test_torch_fov_y
• .readme/equi2pers.md — full parameter documentation and intrinsic matrix explanation
• README.md — mark both TODOs as done