refactor: Rename nodes to n in generators and centr_*_tmax()

[krlmlr]

Part of #2788 (PR 2 of the stack, on top of #2859); executes the nodes → n row of the survey — the direction #692 recorded as essentially uncontested (62 functions already use n).

What this does

Renames the vertex-count argument nodes to n in 16 functions:

• Game generators (11): sample_grg/grg, sample_pref/pref, sample_asym_pref/asym_pref, sample_traits/traits, sample_traits_callaway/traits_callaway, sample_forestfire.
• Centralization (4): centr_betw_tmax, centr_clo_tmax, centr_degree_tmax, centr_eigen_tmax. The latter two had no registry entry and no ... yet, so they gain their standard 3.0.0 keyword-only migration here (loops stays required-with-no-default in the tail, unfold_tree() style; the existing loops/scale lifecycle guards are preserved).
• graph_from_graphdb (1): nodes sits after ..., so this is a pure registry rename (bare-symbol nodes = n in old) — legacy nodes = calls are recovered by the generated block under the usual single soft-deprecation.

nodes is a head formal everywhere else, which the recovery machinery cannot rescue from ... — so those functions follow the as_adjacency_matrix() attr/weights precedent: positional calls bind n unchanged (no warning), and a surviving trailing nodes = deprecated() sentinel catches by-name legacy calls with a soft-deprecation and a conflict error when both spellings are supplied. Dot-case wrapper functions (grg.game() etc.) keep their frozen signatures and forward n = nodes.

Selector-nodes functions (ego() family, constraint(), alpha_centrality(), power_centrality()) are deliberately untouched — they rename to vertices in a later PR of this stack.

Notes for review

• Abbreviated legacy spellings (node =, nod =) of the retired head name now fail loudly with the "spell out the full argument name" guidance instead of partial-matching — same trade-off the attr → weights rename shipped.
• tests/testthat/_snaps/foreign.md was updated by hand for the renamed error text because graph_from_graphdb() needs network access unavailable in the development sandbox; the offline-renderable messages were verified to match exactly.

🤖 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 b391ca2 is merged into claude/issue-2788-subagents-faw1t1:

• ✔️as_adjacency_matrix: 759ms -> 762ms [-0.23%, +1.08%]
• ✔️as_biadjacency_matrix: 765ms -> 765ms [-0.67%, +0.83%]
• ✔️as_data_frame_both: 1.68ms -> 1.66ms [-3.9%, +1%]
• ✔️as_long_data_frame: 4.06ms -> 4.08ms [-1.18%, +2.32%]
• ✔️es_attr_filter: 2.76ms -> 2.73ms [-3.02%, +0.89%]
• ✔️graph_from_adjacency_matrix: 123ms -> 124ms [-0.7%, +2.48%]
• ✔️graph_from_data_frame: 3.71ms -> 3.74ms [-0.44%, +1.98%]
• ✔️vs_attr_filter: 1.57ms -> 1.55ms [-2.55%, +0.98%]
• ✔️vs_by_name: 1.02ms -> 1.03ms [-2.52%, +3.82%]
  Further explanation regarding interpretation and methodology can be found in the documentation.