feat: accept two-column matrices/data frames in edge functions (#1827)

[schochastics]

Summary

Closes #1827. get_edge_ids() was changed in #1663 to accept the natural edge-list forms — an n×2 matrix or a two-column data frame — via the el_to_vec() helper. This PR applies the same treatment to the other functions that take edges as pairs of vertices, so the same edge list can be passed everywhere with no manual transposition (the core complaint in #550):

• add_edges() / delete_edges()
• edge() / edges() (via the + and - operators)
• make_graph() (and make_directed_graph() / make_undirected_graph())
• E(graph, P = )

For all of these, an n×2 matrix or two-column data frame is now the canonical edge-list form. 2×n and 2×2 matrices are deprecated (defunct), consistent with get_edge_ids().

⚠️ Behavior change for make_graph() (please read)

Unlike the other functions, make_graph() already accepted a matrix — but it flattened it column-major. The practical consequences:

• A plain n×2 edge list (e.g. as_edgelist(g)) silently produced the wrong graph.
• The transposed 2×n form (t(edgelist)) was the working idiom.

This PR makes the n×2 matrix the correct, canonical form and turns the 2×n form into a defunct error. This is a breaking change: code relying on make_graph(t(el)) must now pass make_graph(el) (n×2) directly. This is intentional and is exactly what #550 asked for, but it is more disruptive than for the other functions, which previously errored or mis-handled matrices rather than silently accepting a transposed one.

Implementation notes

• el_to_vec() gains arg/fn parameters so error/deprecation messages name the calling function (e.g. "the edges argument of add_edges()"). Defaults preserve the existing get_edge_ids wording, so existing snapshots are unchanged.
• Each entry point gates on is.data.frame(x) || inherits(x, "matrix") before normalizing, so vectors and igraph.vs/igraph.es sequences keep their existing behavior.
• delete_edges() converts a matrix/df of vertex pairs to edge IDs via get_edge_ids(..., error = TRUE) (its edges arg is an edge selector, not a vertex sequence), preserving numeric-edge-id and "a|b" semantics for vectors.
• Internal callers that relied on passing 2×n matrices are updated to pass explicit flat vectors (c(...)): graph_from_edgelist(), graph_from_data_frame(), the incidence/adjacency constructors, and [<- indexing. (graph_from_data_frame() was in fact already broken by the add_edges change and is fixed here.)

Tests

Added matrix/data-frame acceptance, vector/igraph.vs back-compat, 2×n/2×2 rejection, and as_edgelist() round-trip (#550) cases across test-interface.R, test-operators.R, test-make.R, and test-iterators.R. Updated test-conversion.R to use the now-natural n×2 form.

Full suite: 1553 test groups, 0 failures, 0 errors. No snapshot changes.

🤖 Generated with Claude Code

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 713cdd8 is merged into main:

• 🚀as_adjacency_matrix: 760ms -> 730ms [-4.46%, -3.39%]
• 🚀as_biadjacency_matrix: 764ms -> 730ms [-4.76%, -3.94%]
• ❗🐌as_data_frame_both: 1.49ms -> 1.51ms [+0.54%, +2.04%]
• ✔️as_long_data_frame: 3.87ms -> 3.89ms [-0.03%, +1.23%]
• ✔️es_attr_filter: 2.69ms -> 2.69ms [-0.95%, +0.75%]
• 🚀graph_from_adjacency_matrix: 140ms -> 132ms [-8.62%, -2.9%]
• ✔️graph_from_data_frame: 3.42ms -> 3.42ms [-0.78%, +0.88%]
• ✔️vs_attr_filter: 1.53ms -> 1.53ms [-0.76%, +0.96%]
• ✔️vs_by_name: 989µs -> 998µs [0%, +1.95%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if be3d283 is merged into main:

• ✔️as_adjacency_matrix: 832ms -> 828ms [-2.92%, +1.95%]
• ✔️as_biadjacency_matrix: 864ms -> 864ms [-1.2%, +1.22%]
• ✔️as_data_frame_both: 1.7ms -> 1.7ms [-4.53%, +4.35%]
• ✔️as_long_data_frame: 4.44ms -> 4.43ms [-6.13%, +5.93%]
• ✔️es_attr_filter: 3.37ms -> 3.35ms [-7.57%, +6.39%]
• ✔️graph_from_adjacency_matrix: 137ms -> 137ms [-2.52%, +3.63%]
• ✔️graph_from_data_frame: 3.76ms -> 3.82ms [-4.45%, +7.94%]
• ✔️vs_attr_filter: 1.89ms -> 1.87ms [-3.9%, +2.22%]
• ✔️vs_by_name: 1.22ms -> 1.26ms [-1.54%, +7.03%]
  Further explanation regarding interpretation and methodology can be found in the documentation.