Fix A/D tangent basis vectors to canonical (RA, Dec) frame

[matthewholman]

This fixes an error in the A/D vectors that Adam spotted before I could fix it.

a_vec_from_rho_hat and d_vec_from_rho_hat in src/lib/detection.cpp
were producing unnormalized tangent vectors of a y-pole
parameterization (rotated ~70-120° about ρ̂ from the canonical
RA/Dec basis):

A_old = (ρ̂_z, 0, -ρ̂_x) (not unit length)
D_old = (-ρ̂_x ρ̂_y, ρ̂_x²+ρ̂_z², -ρ̂_z ρ̂_y) (not unit length)

These are silent under symmetric σ_RA = σ_Dec — rotation
invariance of the diagonal weight matrix in the LM normal equations
hides the mistake — but bias the fit whenever per-observation RA/Dec
uncertainties differ. The basis is also degenerate at ρ̂_y = ±1
(arbitrary point on the celestial equator) rather than the much less
common ρ̂_z = ±1 (the actual celestial poles).

Replaced with the canonical equatorial-tangent basis:

A = ∂ρ̂/∂α = (-sin α, cos α, 0)
D = ∂ρ̂/∂δ = (-sin δ cos α, -sin δ sin α, cos δ)

Both unit length, mutually orthogonal, orthogonal to ρ̂ everywhere
except at the celestial poles.

Adds tests/layup/test_tangent_basis.py (9 tests) that pin the basis
to the canonical formulas and verify the (ρ̂, A, D) frame is
orthonormal at representative points across the sky.

Dependencies

Stacks on feat/pybind11-eigen-includes (#322), which provides the
pybind11/eigen.h / pybind11/stl.h headers required for the
Python-side tests to access obs.a_vec and obs.d_vec.

Review Checklist for Source Code Changes

• Does pip install still work?
• Have you written a unit test for any new functions? — tests/layup/test_tangent_basis.py
• Do all the units tests run successfully?
• Does Layup run successfully on a test set of input files/databases?
• Have you used black on the files you have updated? — C++ change; N/A