Add kamada_kawai_layout function

[wallies]

Adds rustworkx.kamada_kawai_layout, supporting both PyGraph and PyDiGraph inputs. Implements the original Kamada-Kawai (1989) algorithm: an outer loop selects the node with the largest partial-gradient norm and an inner loop applies a 2D Newton step against the local Hessian until convergence.

Disconnected graphs are laid out by running Kamada-Kawai independently on each connected component and packing the components in a horizontal row, which avoids the visual collapse seen with single-objective Kamada-Kawai on disconnected inputs. Directed graphs have their distance matrix symmetrised before optimising (Kamada-Kawai is fundamentally undirected).

Closes #438

Example outputs

Side-by-side comparisons against networkx.kamada_kawai_layout (NetworkX 3.6.1, which uses L-BFGS-B). Both produce K-K minima; differences are rotation/reflection of the same energy minimum.

Florentine families (15 nodes, connected)

Medici is centrally placed in both. Topologically equivalent under reflection.

3x4 hexagonal lattice (24 nodes, connected)

Both recover the regular hexagonal grid structure.

Three disconnected components: K4, C5, K3

NetworkX 3.6.1 solves K-K globally on the full distance matrix; components overlap when solved jointly (left). This implementation solves each component independently and packs them horizontally (right).

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}