Fractional open-water coverage for the recharge mask and polder DRN

[bdestombe]

Supports NHFLO/models#135: fully fractional handling of open-water cell coverage in 09pwnmodel2.

• riv_from_oppervlakte_pwn stores ds['panden_coverage'] — the panden-covered fraction of each cell, from the intersected piece areas.
• recharge_pond_mask(fractional=True) uses that fraction at panden cells (summed with the lake coverage, clipped at 1) instead of a hard 1.0. When the variable is absent the previous behavior (1.0 at panden RIV cells) is preserved, which keeps the existing contract tests green.
• drn_from_waterboard_data's exclude generalizes from a boolean drop mask to an open-water fraction: the DRN conductance is scaled by the drainable land share 1 - exclude, and the reach is dropped where the land share is zero. Boolean masks scale by 0/1 and behave exactly as before (all 24 pre-existing polder/lakes/panden tests pass unchanged).

The models script combines this with the RWS rws_oppwater_area fraction (and the BGT fraction in the polders_in_detail branch) into one sw_frac, scales RCH and the new EVT package by 1 - sw_frac, and passes it as the polder-DRN exclude — see the companion commit on NHFLO/models#135.

Note: the models notebook CI installs nhflotools@main, so NHFLO/models#135 needs this PR merged first (the old boolean-only ~exclude raises on a float fraction).

Tests: 142 passed, 1 xfailed (3 new: panden coverage known-answer, combined lake+panden fraction with clipping, land-share-scaled DRN conductance).

[bdestombe]

The rch an evt should only be subtracted from the lakes, because they have their own rainfall and evaporation