Add FFT-free spatial sweep alternative for wind shear perturbation (ustars/ustarn)

[Copilot]

The WindShear.compute_shear method computed dtaux/dtauy (Weng et al. perturbation theory, driving ustars/ustarn) exclusively via 2-D FFT. This adds a loop-and-sweep alternative that avoids FFT entirely, structured analogously to the directional Ct sweep in utils.py.

Core algorithm (shear.py)

• New compute_shear_spatial method: applies a Gaussian spatial filter (replaces FFT-based high-frequency filter), computes central-difference bed slopes, then delegates to a row-by-row Cauchy sweep.
• New module-level @njit(cache=True) helper _cauchy_hilbert_sweep: for each y-row sweeps the antisymmetric kernel 1/(k·π) over all nx cells — the discrete Hilbert transform of the slope in the wind direction. Numerically ~88–90 % correlated with the FFT result.
• Amplitude coefficients derived from Weng et al. inner-layer parameters (alpha_s = 2·log(l/z₀)/ul², alpha_n = 2/ul²), consistent with compute_shear.
• Complexity is O(ny · nx²) vs FFT's O(ny · nx · log nx); a performance note is included in the docstring.

Method selection

WindShear.__init__ gains a method parameter ('fft' | 'spatial', default 'fft'). Existing behaviour is unchanged.

# Select the new sweep-based method in model config:
# method_shear = 'spatial'

# Or directly:
ws = WindShear(x, y, z, dx=dx, dy=dy, L=L, l=l, z0=z0,
               buffer_width=buf, method='spatial')

wind.initialize and constants.py updated to recognise 'spatial' as a valid method_shear value alongside 'fft', 'quasi2d', and 'duna2d'.

Tests (tests/test_shear.py)

12 new tests covering _cauchy_hilbert_sweep (flat bed → zero, antisymmetric hill response, amplitude linearity, row independence) and WindShear with method='spatial' (zero-wind, non-trivial output, ≥ 0.7 correlation with FFT, shape, backward compatibility of FFT path).

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.