add a default warpper

[haleqiu]

Warp-accelerated drop-in replacement for PyPose Lie operations.

# Warp-accelerated (default for most code)
import PyposeWarp as pp
R = pp.SO3(torch.tensor([0., 0., 0., 1.]))  # warp-backed
w = pp.so3(torch.zeros(3)).Exp()             # warp Exp kernel
T = pp.identity_SE3(batch_size)              # warp-backed identity

# Generic pypose fallback (for modules not yet supported, e.g. EKF Jacobians)
import PyposeWarp.generic as pp
R = pp.SO3(torch.tensor([0., 0., 0., 1.]))  # original pypose
J = torch.autograd.functional.jacobian(f, x, vectorize=True)  # works

Both can coexist in the same process. PyposeWarp patches pypose globally
on import, including pypose.module.IMUPreintegrator internals. The generic
submodule provides a snapshot of the original constructors for code that needs
autograd Jacobian compatibility.

[MarkChenYutian]

A bit concerned about this - the export pipeline seems too complicated.