Add paper-derived conformity tests for the supplement equations

[soaressgabriel]

Description

The supplement of the RUBEM paper (equations S1 to S35, PDF pages 5 to 12) is the normative description of the model, but no test derives its expectations from those equations: the unit tests pin values computed by the code itself and the goldens pin the synthetic run. After the review of the equations against the code (fixes #318, #319, #320, #322; confirmed rules in #331), the model needs tests that fail when an implementation drifts from the formulation.

Proposed solution

A paper marker and a tests/paper package with:

• equation tests on a 1x1 clone for the process functions (S3, S5 to S12, S14 to S20, S22, S23, S25 to S29, S31 to S33): expectations computed in float64 from the supplement formulas with explicit tolerances, each test citing the equation and its page;
• model-level tests on the synthetic dataset with rewritten lookup tables (a_o = 1, a_i = 1, saturated start) for the rules the model applies inline: S1 (zero floor and water cell), S2, S4 saturation, S13/S28 water-cell guards, S21/S30 impervious constant, S24 threshold, S34 and S35 (accumulation reproduced with accuflux at x = 0);
• expectations following the corrected code of Fix units soil moisture equation for ch #321, Avoid 1e-5 added to precipitation for all timesteps #323, Avoid kc becomes zero when NDVI == 1.1 * NDVI_min #324 and Avoid negative saturated-zone storage by limiting baseflow to available water #325, so the branch is stacked on those pull requests.

No xfail is needed: every equation is either implemented as written or covered by a confirmed rule (#331).

Alternative solutions

Property-based tests over random inputs were considered; fixed analytic cases keep every derivation readable and citeable.

Additional context

These tests guard the architecture refactors that follow them.