surtgis v0.10.0

Headline: closes the fluvial-tectonic morphometry spec —
docs/SPEC_morfometria_fluvial_tectonica.md — that Dr. Paulo Quezada
(LAMIR/UFPR + UDD) contributed for the Frente Puerto Aysén project.
Adds the five canonical metrics of tectonic geomorphology as a new
crates/algorithms/src/fluvial/ submodule, all reachable from a single
binary as surtgis fluvial <sub>.

Added — crates/algorithms/src/fluvial/

• chi-transform (Perron & Royden 2013): base-level reference
  distance χ, the path integral of (A₀/A(x))^θref. BFS upstream from
  every outlet on a StreamGraph built from the binary stream raster +
  D8 flow_dir. Riemann-sum convention matching TopoToolbox 2 for
  bit-for-bit parity.
• channel-steepness (ksn) (Wobus 2006): channel-following slope
  S = (z_up − z_down)/dx_along_channel, raw ksn = S · A^θref,
  smoothed over a moving window (default 500 m) along the network with
  main-stem traversal at confluences. Optional --segments flag emits a
  LineString GeoJSON of per-segment averages.
• knickpoints (Neely 2017): per-segment χ–z profile + 1-D TVD
  denoising (inline Condat 2013, ~80 LOC; the tvr crate that the spec
  recommended doesn't exist on crates.io) + non-uniform 3-point
  curvature stencil + magnitude / polarity classification (concave =
  decreasing slope downstream → likely lithology; convex = increasing
  slope downstream → likely transient/tectonic). Confluence buffer to
  suppress edge artefacts. GeoJSON Points + optional categorical raster.
• concavity (Perron & Royden 2013): per-basin θ estimation via
  grid search minimising elevation~χ scatter, with deterministic seeded
  bootstrap (default n=200) for 95 % CI. CSV output with theta_opt,
  theta_ci_low, theta_ci_high, n_cells, rmse.
• divide-migration (Willett 2014, Whipple 2017): scan 4-connected
  basin boundaries, group by sorted basin-id pair, compute median Δχ +
  Gilbert Δelev + Δrelief (local 3×3 max−min). Greedy nearest-neighbour
  polyline geometry per divide. LineString GeoJSON output.

Foundation primitive used by all five: StreamGraph +
build_stream_graph(stream, flow_dir) in fluvial/stream_traversal.rs.

29 unit tests, including the five §7.2 headline golden tests from
the spec (all pass).

Added — CLI

Five new subcommands under surtgis fluvial:

• fluvial chi STREAM FLOW_DIR FLOW_ACC OUTPUT [--theta-ref 0.45] [--a-0-m2 1e6]
• fluvial ksn STREAM FLOW_DIR FLOW_ACC DEM OUTPUT [--segment-length-m 500] [--segments GEOJSON]
• fluvial knickpoints STREAM FLOW_DIR FLOW_ACC DEM OUTPUT [--tvd-lambda 0.5] [--raster RASTER]
• fluvial concavity STREAM FLOW_DIR FLOW_ACC DEM BASINS OUTPUT [--bootstrap-n 200]
• fluvial divide-migration BASINS DEM FLOW_ACC OUTPUT [--chi CHI] [--min-divide-length-m 500]

All handlers share a CRS-validation heuristic: reject inputs with
pixel size < 1.0 unit (likely degrees) unless --cell-size-m is
supplied explicitly. Warns when cell size > 30 m per spec §8 pitfall
#1 (ksn / knickpoint sensitivity).

Added — hydrology stream-network --from-facc

Optional flag to interpret the input as a pre-computed flow_accumulation
raster (skipping the DEM-side recomputation). Required for composable
workflows where stream-network must be topologically consistent with an
externally-computed flow_dir. Backward-compatible (default behaviour
unchanged).

Notes

• Versioning is minor (0.9 → 0.10) because all additions are
  non-breaking; the existing public API surface is untouched. No
  Breaking heading required this release.
• Inter-crate path-deps bumped 0.9 → 0.10 across all 8 crates.
• Sprints 7-9 of the spec (TopoToolbox / pyTopoToolbox parity
  validation against published Smugglers Notch case + the Quezada
  Pangal AOI cross-check + mdBook chapter) are deferred to a follow-up
  release. They validate the algorithms against the literature but do
  not block usability for early adopters.