Fix inconsistent shift_operator (#146)

[astanziola]

Fixes #146

The FiniteDifferences shift_operator used dx[0] for all axes instead of dx[i] per axis, and did not broadcast single-element dx lists to multi-dimensional domains (unlike FourierSeries).

Changes

• FiniteDifferences shift now applies dx[i] per axis
• Single-element dx is broadcast to all dimensions for both discretizations
• Updated docstrings for dx argument

Tests

• Pinned exact numerical outputs for 1D shifts (both FD and Fourier)
• Per-axis dx verification for 2D vector fields
• Broadcast equivalence (dx=[a] == dx=[a, a])
• Direction consistency between FD and Fourier
• Zero-shift identity
• jit compatibility

Closes #146