refactor: snake_case attr-comb options & arguments, soft-deprecate dotted names

[schochastics]

Split out from #2676 per review feedback ("I'd like to review the par code in a separate PR"). #2676 kept the original dotted option/argument names; this PR carries the naming refactor on its own so it can be reviewed in isolation.

What this does

• igraph options: rename vertex.attr.comb → vertex_attr_combine, edge.attr.comb → edge_attr_combine, and the new graph.attr.comb → graph_attr_combine. Reading/setting the old dotted keys still works via back-compatible aliasing in igraph_opt() / igraph_options(), emitting a single lifecycle::deprecate_soft() warning (R/par.R).
• Function arguments: simplify(), as_undirected() and contract() gain snake_case edge_attr_combine / vertex_attr_combine arguments via the argument-migration infrastructure (registry entries in tools/migrations/attr-comb.R, spliced by Rscript tools/generate-migrations.R; schema in tools/migrations/README.md); the dotted names keep working and soft-deprecate.
• Graph operators: union() / intersection() / compose() / disjoint_union() use the snake_case *_combine argument names.
• Updates the stimulus codegen defaults (types-RR.yaml), regenerated *_impl defaults, and the affected tests/snapshots.
• Points the stale tools/migrations.R schema references in sibling registry files and helper comments at tools/migrations/README.md, where the schema documentation lives since the registry split (refactor(migrations): split the registry into per-topic files #2779).

Base branch

#2676 has merged, so this PR is based on main and rebased into a single commit on top of it; the attr-comb migration entries live in their own registry file tools/migrations/attr-comb.R so that the open per-topic migration PRs can append to conversion.R, community.R, etc. without conflicting here.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 1453158 is merged into feat-attrib_comb:

• ✔️as_adjacency_matrix: 706ms -> 701ms [-2.32%, +0.95%]
• ✔️as_biadjacency_matrix: 734ms -> 736ms [-1.27%, +1.75%]
• ✔️as_data_frame_both: 1.48ms -> 1.56ms [-8.29%, +19.78%]
• ✔️as_long_data_frame: 3.85ms -> 3.91ms [-1.87%, +5.16%]
• ✔️es_attr_filter: 2.84ms -> 2.88ms [-1.61%, +4.71%]
• ✔️graph_from_adjacency_matrix: 147ms -> 148ms [-1.36%, +2.48%]
• ✔️graph_from_data_frame: 3.38ms -> 3.44ms [-1.16%, +4.6%]
• ✔️vs_attr_filter: 1.54ms -> 1.56ms [-3.55%, +5.83%]
• ✔️vs_by_name: 983µs -> 983µs [-3.85%, +3.81%]
  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 2f878f3 is merged into main:

• ✔️as_adjacency_matrix: 924ms -> 934ms [-1.4%, +3.52%]
• ✔️as_biadjacency_matrix: 899ms -> 899ms [-1.57%, +1.56%]
• ✔️as_data_frame_both: 1.68ms -> 1.67ms [-3.86%, +2.57%]
• ✔️as_long_data_frame: 4.38ms -> 4.22ms [-13.07%, +5.49%]
• ✔️es_attr_filter: 3.16ms -> 3.09ms [-5.92%, +1.93%]
• ✔️graph_from_adjacency_matrix: 164ms -> 168ms [-1.87%, +6.71%]
• ❗🐌graph_from_data_frame: 3.63ms -> 3.77ms [+1.29%, +6.67%]
• ✔️vs_attr_filter: 1.84ms -> 1.81ms [-7.05%, +3.86%]
• ✔️vs_by_name: 1.11ms -> 1.11ms [-4.04%, +3.9%]
  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 ba22989 is merged into main:

• ✔️as_adjacency_matrix: 760ms -> 758ms [-0.87%, +0.2%]
• ✔️as_biadjacency_matrix: 742ms -> 740ms [-0.98%, +0.45%]
• ✔️as_data_frame_both: 1.46ms -> 1.47ms [-1.02%, +1.2%]
• ✔️as_long_data_frame: 3.87ms -> 3.87ms [-0.66%, +1.04%]
• ✔️es_attr_filter: 2.77ms -> 2.75ms [-2.88%, +1.57%]
• ✔️graph_from_adjacency_matrix: 118ms -> 117ms [-1.93%, +1.58%]
• ❗🐌graph_from_data_frame: 3.32ms -> 3.38ms [+1.01%, +2.53%]
• ✔️vs_attr_filter: 1.5ms -> 1.5ms [-1.86%, +0.59%]
• ✔️vs_by_name: 1ms -> 987µs [-4.57%, +1.83%]
  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 c3cf816 is merged into main:

• ✔️as_adjacency_matrix: 753ms -> 753ms [-1.42%, +1.22%]
• ✔️as_biadjacency_matrix: 723ms -> 724ms [-0.92%, +1.07%]
• ✔️as_data_frame_both: 1.54ms -> 1.53ms [-2.27%, +0.28%]
• ✔️as_long_data_frame: 3.85ms -> 3.85ms [-0.94%, +0.93%]
• ✔️es_attr_filter: 2.81ms -> 2.8ms [-1.78%, +1.06%]
• ✔️graph_from_adjacency_matrix: 143ms -> 142ms [-1.65%, +1.25%]
• ❗🐌graph_from_data_frame: 3.44ms -> 3.5ms [+0.51%, +3.2%]
• ✔️vs_attr_filter: 1.52ms -> 1.51ms [-2.31%, +1.06%]
• ✔️vs_by_name: 1ms -> 988µs [-3.25%, +0.42%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[krlmlr]

Thanks! Almost good to merge. The one snapshot flagged could be just a symptom of a pattern across the codebase.

[krlmlr]

This doesn't look quite right, is this our code that triggers this deprecation message?