Fix Maritorena and Stumpf depth-correction formulas

[ysims]

_maritorena_depth was previously a single-parameter exponential radiance scaler with no relation to any published Maritorena model. Replace it with the inverse of the Maritorena, Morel & Gentili (1994) two-flow shallow-water reflectance model: given known depth, deep-water reflectance (Rw_inf), and diffuse attenuation (K_d, repurposing the existing attenuation parameter), recover bottom reflectance Rb from measured water-leaving reflectance Rw via Rb = Rw_inf + (Rw - Rw_inf) / exp(-2 * K_d * depth), the algebraic inverse of the forward model Rw = Rw_inf + (Rb - Rw_inf) * exp(-2 * K_d * depth). Both Rw_inf and K_d now accept per-band mappings via the existing _band_param helper, matching the Lyzenga functions' convention.

_stumpf_depth previously computed m0 - m1 * ln(blue) / ln(green) directly on raw reflectance, which both flips the sign convention of Stumpf, Holderied & Sinclair (2003) eq. (1) (m1 * ratio - m0, not m0 - m1 * ratio) and omits the paper's fixed scaling constant n (commonly 1000), which keeps ln(n * Rw) positive and away from the numerical instability of taking logs of small fractional reflectance values near zero. This is a genuine formula bug, not just a re-parameterisation: fix both the sign and add n (default 1000.0, matching the original paper) as a keyword.

Behavioural change: existing callers of method="maritorena" or method="stumpf" with default parameters will see different numeric output, which is expected -- the old formulas did not implement the cited papers.