Stabilize DESC radial evaluation and prune zero Fourier modes

[paulromano]

This PR makes three small improvements:

• Evaluate Zernike radial polynomials using SciPy's Jacobi polynomials instead of the alternating factorial sum, avoiding cancellation at high radial orders
• Drop converted Fourier modes whose R and Z coefficients are exactly zero on every radial surface. This preserves the represented geometry while reducing setup and sampling work. I've run into some cases where the highest order l/m can be all zeros, so this can actually help reduce the maximum order in that case.
• Restore unrelated settings documentation that was inadvertently reverted on the branch

[eepeterson]

LGTM!