Add Fosberg dead fuel moisture grid SDK support

[amarcozzi]

Summary

Wraps the generated create_fosberg_fuel_moisture_grid endpoint as a new grid family (fuel_moisture) with no prior SDK surface:

• Adds ff.grids.create_dead_fuel_moisture_grid_from_fosberg(...), returning a pending Grid whose single fuel_moisture.dead.1hr band is derived from a topography grid (slope + aspect) and a Leaflux irradiance grid (irradiance.surface.relative).
• Source grids are accepted as Grid objects or bare id strings. When a Grid is passed, it is checked for completion and required bands, and the domain id is resolved from it; when both sources are bare ids the call raises a clear ValueError (the domain can't be resolved).
• Validates the ranges the request model documents — dry_bulb_temp >= 10, relative_humidity 0–100, time in HHMM 0800–1959 (minutes 00–59) — mirroring the _duet_integer style, and coerces month/elevation to their enums.
• Exports the function via grids.__all__ (module-qualified ff.grids.create_*, matching the other creators).

Testing

• Added TestCreateDeadFuelMoistureGridFromFosberg mirroring the DUET test class: request-building (via monkeypatched endpoint), grid-object/id resolution, domain-resolution error, completion/band guards, and parametrized range validation. 16 offline unit tests pass; full offline unit selection (Fosberg + helpers + alignment + signatures) is 35 passing. black and flake8 are clean (enforced by pre-commit).
• Added a live end-to-end test (test_create_live) that voxelizes a tree inventory (leaf_area_density), builds a Leaflux irradiance grid, then creates and waits on the Fosberg grid.

Live-test status

The live API and credentials work (the voxelize step runs live). However the generated Leaflux client is currently out of sync with the live server: it sends source_grid_id while the server now requires source_lad_grid_id, so the irradiance prerequisite can't be built yet. This is a Leaflux (#195) client-regeneration concern, not part of the Fosberg endpoint. The live test therefore pytest.skips with a clear reason when the irradiance grid can't be created, so it exercises the Fosberg call end-to-end once the Leaflux client is realigned, without spuriously failing CI in the meantime.

Closes #197

[amarcozzi]

On hold — do not merge yet.

The Fosberg SDK wrapper and its offline tests are complete and passing, and this branch has been merged up to date with main (picking up the leaflux source_lad_grid_id rename). However, the live end-to-end test cannot pass until an upstream backend bug is fixed:

• The Fosberg grid consumes an irradiance.surface.relative leaflux grid.
• That surface-irradiance computation currently fails server-side in treevox when the terrain grid's dimensions differ from the LAD grid: ValueError: Terrain dimensions must match leaf area grid dimensions.
• Tracked in Leaflux surface irradiance fails when terrain grid dimensions differ from the LAD grid FastFuels-API-v2#547.

Holding this PR until #547 lands, after which test_create_live should pass without changes.

[amarcozzi]

Unblocked — ready to merge.

FastFuels-API-v2 #548 (reject source grids that don't share a horizontal lattice) is merged and deployed to api-v2-prod (revision api-v2-prod-00232-d4w).

Merged main into this branch and updated the Fosberg live test for the new contract: the topography grid is now built with create_topography_grid_from_3dep(..., align_to=voxels, bands=["elevation", "slope", "aspect"]) so the topography, LAD, and leaflux irradiance grids all share one horizontal lattice (elevation feeds the leaflux surface draping; slope/aspect feed Fosberg). test_create_live now runs the full chain (voxelize → aligned topo → surface irradiance → Fosberg) and passes end-to-end against the deployed API (verified, 52s). Offline unit tests pass; black/flake8 clean.