Add StarGraph[n] and fix the generator argument-array leak

[msollami]

Summary

Adds StarGraph[n] alongside the existing basic graph constructors, and fixes an
Expr** argument-array leak that the new function surfaced by reproducing it — the
other four generators in src/graph/generators.c had it too.

StarGraph[n] gives the star on vertices 1..n: the hub 1 joined to each of the
n-1 leaves 2..n, undirected. Hub vertex 1 matches Mathematica and the
int_vertices() convention the file's other generators already use.

Changes

• src/graph/generators.c — builtin_star_graph, modelled on builtin_cycle_graph.
  n = 1 is a lone vertex, n = 0 the empty graph; a non-integer or negative
  argument leaves the expression unevaluated, as with the siblings. No wrap-edge
  special case is needed, unlike CycleGraph[2], because a star has no duplicate
  edge at any size.
• src/sym_names.{h,c} — SYM_StarGraph interned.
• src/graph/graph.h — builtin_star_graph declared.
• src/graph/graph.c — registered in graph_init with Protected and a docstring.
• tests/test_graph.c — ten assertions in test_generators.
• docs/spec/builtins/graphs.md, docs/spec/changelog/2026-08-31.md,
  Mathilda_spec.md — reference, changelog entry, changelog table row.

The leak fix

expr_new_function() memcpy's the argument array rather than taking ownership of
it, so the calloc'd Expr** is the caller's to free — and of the five generators
only RandomGraph did. The shared make_graph helper now owns both arrays via a
new make_list_owning(), which covers CompleteGraph, CycleGraph, PathGraph
and StarGraph in one place; RandomGraph's inline vertex list uses it too.

leaks -atExit -- ./graph_tests drops from 715 records / 40944 bytes to
640 / 36816, with no generator frames remaining apart from RandomGraph's
separate RandomSample path (3 records), which is pre-existing and untouched here.

Testing

tests/build/graph_tests — all 17 cases pass, including the new test_generators
assertions.

Verified against the built binary:

{VertexCount[StarGraph[5]], EdgeCount[StarGraph[5]]}   (* {5, 4}                *)
EdgeList[StarGraph[4]]        (* {1 <-> 2, 1 <-> 3, 1 <-> 4}                    *)
VertexDegree[StarGraph[5]]    (* {4, 1, 1, 1, 1}                                *)
DirectedGraphQ[StarGraph[5]]  (* False                                          *)
AdjacencyMatrix[StarGraph[4]] (* {{0,1,1,1},{1,0,0,0},{1,0,0,0},{1,0,0,0}}      *)
{VertexCount[StarGraph[1]], EdgeCount[StarGraph[1]]}   (* {1, 0}                *)
{VertexCount[StarGraph[0]], EdgeCount[StarGraph[0]]}   (* {0, 0}                *)
ConnectedGraphQ[StarGraph[6]]                (* True — a star is its own tree   *)
EdgeCount[FindSpanningTree[StarGraph[6]]]    (* 5                               *)
Count[GraphPlot[StarGraph[5]], _Line, Infinity]        (* 4                     *)
Head[StarGraph[x]]  /  Head[StarGraph[-3]]   (* StarGraph — both unevaluated    *)

Also green: make (clean, -Wall -Wextra), make check-c99,
make check-packed-aware, make check-nd-surfaces.

Not run: the full tests/build suite — it exceeded the time budget locally, so
only graph_tests was executed. Nothing in this diff touches non-graph code, but
that is an argument rather than a measurement. No packed/NDArray or Compile[]
surfaces apply: this is a structural constructor returning a Graph[...]
expression, not a numeric head.

JIRA Ticket

n/a