refactor: Rename vertex-selector arguments to vertices

[krlmlr]

Part of #2788 (PR 6 of the stack, on top of #2863); executes the selector decision along the #692 scheme — the survey's most recent point of near-consensus (szhorvat/ntamas/maelle/krlmlr in #692 and #853, maelle again in the #2788 thread): vertices for a subset, n for a count (PR 2 of this stack), v for a single vertex.

What this does

Renames the vertex-subset selector to vertices in 24 functions (+3 aliases that follow automatically):

• Head selectors, registry + sentinel (positional calls unchanged and silent; named legacy calls soft-deprecated via a surviving trailing deprecated() sentinel): betweenness, closeness, strength, harmonic_centrality, power_centrality, alpha_centrality (was v/vids/nodes); degree, distances (to untouched), constraint, ego, ego_size, make_ego_graph, induced_subgraph, knn; adjacent_vertices, incident_edges; local_efficiency, similarity.
• Tail selectors, pure registry rename: diversity, page_rank, transitivity.
• Functions without registry entries (sentinel + bridge, no ... inserted): cocitation, bibcoupling, delete_vertices, isomorphism_class, count_triangles, max_degree.

Single-vertex arguments deliberately keep their names per #692: neighbors(v), incident(v), subcomponent(v), cluster_spinglass(vertex).

Abbreviation guards regenerate accordingly: v/vi/vid on the vids functions and v–verte on head cases become explicit ambiguity/spell-out errors where they could mean either spelling; a bare v = on former-vids functions now silently partial-matches vertices (same meaning it always had).

Full test suite: 9357 passing; the only failures are the two environment-bound ones (network for graph_from_graphdb, callr needing an installed igraph).

Deliberately left for follow-up

• eccentricity(vids = V(graph)) — hand-rolled dots handling and a non-constant default the generator rejects; needs its real migration first.
• subgraph() — docs promise subgraph_from_edges() takes over this name in the next major; renaming its vids now would churn twice. (Shared Rd note: its still-live vids formal shows the deprecated-badge text from induced_subgraph's page.)
• graph.isoclass.subgraph() (undocumented legacy alias), shape_noplot() (its v is part of the documented vertex-shape API contract).
• The edge mirror (eids/es/e) — explicitly an open question in Comprehensive plan: argument renames for 3.0.0 #2788 (edges is taken by 14 constructors, though no function carries both, so edges is mechanically feasible).
• index in the 11 attribute accessors — spans vertex and edge subsets, so it cannot follow this rename without a per-type split.
• distances(vertices = , to = ) vs the from/to family — the naming tension flagged in Comprehensive plan: argument renames for 3.0.0 #2788 remains a separate decision.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RTPj4qNv2FWui6etixZeuR

---

Generated by Claude Code

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 7954b20 is merged into claude/issue-2788-subagents-faw1t1-5:

• ✔️as_adjacency_matrix: 775ms -> 778ms [-1.1%, +2.03%]
• ✔️as_biadjacency_matrix: 780ms -> 773ms [-2.39%, +0.75%]
• ✔️as_data_frame_both: 1.63ms -> 1.63ms [-1.48%, +2.41%]
• ✔️as_long_data_frame: 4ms -> 4.01ms [-0.44%, +0.97%]
• ✔️es_attr_filter: 2.74ms -> 2.76ms [-2%, +3.32%]
• ✔️graph_from_adjacency_matrix: 128ms -> 127ms [-2.09%, +0.89%]
• ✔️graph_from_data_frame: 3.68ms -> 3.68ms [-0.72%, +0.9%]
• ✔️vs_attr_filter: 1.61ms -> 1.52ms [-13.35%, +2.55%]
• ✔️vs_by_name: 1.03ms -> 1.04ms [-1%, +1.95%]
  Further explanation regarding interpretation and methodology can be found in the documentation.