feat: QHH baseline Step1-3 regression pass

[DankerMu]

Summary

Closes #21. QHH baseline end-to-end regression verification (Step1-Step3) with the sf/terra spatial stack.

Changes

• AutoSHUD submodule updated to dbf7b6c (PRs [T2] tools/shudnc.py: enable nc profile validate/run #6-[OpenSpec] Mark merged changes complete and archive (A/B migration) #10: complete sf/terra migration, rgdal/rgeos removal)
• docs/regression-qhh.md: regression record with environment, artifacts, statistics, and known differences

Verification Results

Metric	Value
Mesh cells	4812
River segments	1633
Watershed area	29613.7 km²
Forcing stations	386
Time range	2017-2018

Known Differences

All differences are cosmetic or negligible:

1. 4 empty geometries filtered from stream shapefile (sf preserves them, sp cannot)
2. Length computation uses sp::SpatialLinesLengths instead of rgeos::gLength (< 1e-10 difference)
3. Spatial intersection via sf::st_intersects instead of rgeos::gIntersects (same GEOS backend)

AutoSHUD PRs

• [T2] tools/shudnc.py: enable nc profile validate/run #6: drop empty geometries in read_sf_as_sp
• [R] tools: sampled regression compare #7: replace readOGR/gIntersects with sf equivalents
• [Docs] Align Phase A Forcing SPEC with implemented cfg.forcing keys (LAYOUT_*/NC_*) #8: terra API in raster2Polygon
• [Docs] Align Phase B Output SPEC with implemented cfg.ncoutput keys (SCHEMA/OUT_DIR) #9: handle empty CRS from xyz2Raster
• [OpenSpec] Mark merged changes complete and archive (A/B migration) #10: complete sf/terra migration for Step2+Step3

[DankerMu]

• CI: OpenSpec Validate + Python Smoke passed.
• AutoSHUD submodule: updated to dbf7b6c and it matches DankerMu/AutoSHUD master HEAD.
• docs/regression-qhh.md: includes environment, run commands, key artifacts, summary statistics, and known differences.
• Diff contains only the submodule gitlink + docs; no secrets or large data.
• PR commit references issue [SHUD-NC] 跑通 QHH baseline（sf/terra 版本）并记录回归对比/容差 #21.