feat: graph_from_data_frame(vertex_ids = TRUE) reads numeric vertex IDs, so as_data_frame(what = "both") round-trips unnamed graphs

[schochastics]

Closes #223.

Until now, graph_from_data_frame() always read vertices[, 1] as vertex names and converted the endpoints to strings. That meant as_data_frame(g, what = "both") could not be passed back in for graphs without vertex names. This PR adds an opt-in keyword argument:

df <- as_data_frame(g, what = "both")
graph_from_data_frame(df$edges, vertices = df$vertices, vertex_ids = TRUE)

With vertex_ids = TRUE:

• The first two columns of d must be positive whole numbers. They are used as vertex IDs.
• Every column of vertices becomes a vertex attribute, one row per vertex in ID order. No name attribute is created unless vertices has a name column.
• If vertices is NULL, the vertex count is the largest ID in d, so isolated vertices up to that ID are kept.

The default (vertex_ids = FALSE) behaves as before. In addition:

• The "Some vertex names in d are not listed in vertices" error now suggests vertex_ids = TRUE when the endpoints are in-range integers.
• A zero-column vertices used to fail with an internal [.data.frame error. It now reports "vertices contains no columns" with the same hint. as_data_frame() returns exactly this for an unnamed graph without vertex attributes.

Notes:

• vertex_ids is also added to the new signature in tools/migrations/conversion.R, and the ARG_HANDLE block was regenerated. As a result, the abbreviations v, ve, ver and vert now abort as ambiguous. Before, they were recovered as vertices with a deprecation warning.
• Graph-level attributes (such as make_ring()'s name) are not stored in the data frames, so the round trip is exact only for vertex and edge data.

🤖 Generated with Claude Code

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if dde9886 is merged into main:

• ✔️as_adjacency_matrix: 127ms -> 127ms [-1.13%, +1.8%]
• ✔️as_biadjacency_matrix: 136ms -> 136ms [-1.89%, +1.48%]
• ✔️as_data_frame_both: 146ms -> 146ms [-0.72%, +1.75%]
• ✔️as_long_data_frame: 119ms -> 120ms [-0.58%, +2.9%]
• ✔️es_attr_assign: 127ms -> 128ms [-0.99%, +2.53%]
• ✔️es_attr_filter: 137ms -> 137ms [-0.65%, +0.82%]
• ✔️graph_from_adjacency_matrix: 161ms -> 161ms [-1.14%, +1.65%]
• ✔️graph_from_data_frame: 159ms -> 160ms [-0.51%, +1.75%]
• ✔️incident_edges: 143ms -> 142ms [-3.07%, +1.82%]
• ✔️neighbors_by_name: 140ms -> 140ms [-1.73%, +1.21%]
• ✔️union_named: 138ms -> 139ms [-1.86%, +2.58%]
• ✔️vs_attr_filter: 166ms -> 167ms [-0.25%, +0.74%]
• ✔️vs_by_name: 147ms -> 147ms [-0.17%, +1.01%]
  Further explanation regarding interpretation and methodology can be found in the documentation.