ggrust 0.8.0

Added

• geom_streamline() / stat_streamline(), as R metR: ggplot(data) + geom_streamline(aes(x=, y=, u=, v=)) draws streamlines with one arrowhead half way along each line, per group and facet panel. Rows that form a complete regular grid are used as the field; scattered rows are first averaged onto a grid with velocity_grid() (gridded= forces either). Lines are integrated with streamplot() (density, minlength, maxlength, integration_direction, broken_streamlines). New stat-only aesthetics u and v. Not in lets-plot (see deviations.md).
• velocity_grid(points, vectors, ...): average vectors at scattered points onto a regular grid (Gaussian-weighted nearest neighbours), as scVelo's compute_velocity_on_grid, without scVelo. Returns a VelocityGrid(x, y, u, v) that feeds streamplot() directly. Same values as scVelo 0.3.4 on the reference cases; about 7x faster.
• density_outline(points, level=...): outline a 2D point cloud where its Gaussian kernel density falls to level times its peak, as scipy's gaussian_kde plus scikit-image's find_contours, without either. Returns an (x, y, path) frame for geom_path(). Same contours as scipy 1.17 / scikit-image 0.26 on the reference cases; about 10x faster.
• kernel_density(values, at, bandwidth=...): the Gaussian kernel density of a 1D sample at given points, as scipy's 1D gaussian_kde (Scott, Silverman or a factor), returned as an (x, density) frame. Within 1e-13 of scipy's values on the reference cases.