ci: Sync R-CMD-check workflow; drop matrix-env plumbing from custom action

[krlmlr]

This PR was prepared with Claude Code (LLM disclosure).

• By submitting this pull request, I assign the copyright of my contribution to The igraph development team.

---

Sync .github/workflows/R-CMD-check.yaml and
.github/workflows/versions-matrix/action.R from cynkratemplate
(cynkra/cynkratemplate#95).

The claim from #2782 —
that a custom before-install action is needed
to pick up the env field of a test matrix entry —
was accurate under the old template design,
and it applied to all template consumers, not just rigraph:
composite actions cannot read the matrix context,
so the workflow forwarded matrix.env as an action input,
and each repo had to apply it to $GITHUB_ENV itself.

The template's rcc-full job now applies the generic env field
to $GITHUB_ENV directly, before the custom before-install action runs,
so no repo needs plumbing of its own.
The apply boilerplate and the env input
are therefore removed from the custom before-install action;
IGRAPH_LIFECYCLE_ERRORS=true from the lifecycle-errors matrix entry
keeps reaching the tests through the workflow-level step.

Additionally, this PR regenerates the ARG_HANDLE blocks
(Rscript tools/generate-migrations.R, generated code only):
current main carries drift —
most likely from the merge order of today's migration refactors
(#2760, #2761, #2762 and the registry side of #2782) —
and the rcc-smoke drift check fails on every branch
until the regenerated blocks are committed.
The first CI run on this PR failed exactly there,
before the regeneration commit was added.

🤖 Generated with Claude Code

https://claude.ai/code/session_0136C9bRMjBGkUrtLdkYNmAg

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if fd5b37c is merged into main:

• ✔️as_adjacency_matrix: 793ms -> 792ms [-1.36%, +1.06%]
• ✔️as_biadjacency_matrix: 771ms -> 775ms [-1.28%, +2.3%]
• ✔️as_data_frame_both: 1.6ms -> 1.57ms [-4.41%, +1.54%]
• ✔️as_long_data_frame: 4.01ms -> 3.97ms [-2.28%, +0.46%]
• ✔️es_attr_filter: 2.78ms -> 2.75ms [-4.53%, +1.88%]
• ✔️graph_from_adjacency_matrix: 126ms -> 125ms [-3.5%, +1.12%]
• ✔️graph_from_data_frame: 3.45ms -> 3.6ms [-2.85%, +11.25%]
• ✔️vs_attr_filter: 1.57ms -> 1.62ms [-3.07%, +9.01%]
• ✔️vs_by_name: 974µs -> 968µs [-2.48%, +1.14%]
  Further explanation regarding interpretation and methodology can be found in the documentation.