chore: Re-run the reverse-dependency check (most, depth 2)

[krlmlr]

Prepared with Claude Code (LLM-assisted).

Refreshes the committed revdep/ record from the first dispatch-triggered revdep4 run (33777134786, most, depth 2, R 4.6.1) — the first run under the core-cap and timeout-floor fixes of #2875. All 25 jobs green, ~6.1 h wall clock; the universe delta-rebuilt from :latest in 56 min.

Result: 3436 ok · 26 newly broken · 5 timeouts · 1 failed · 1 error · 1 deferred, of 3470 reverse dependencies (the enumeration grew by 29 in the 12 days since the previous run).

Compared with the previous record (run 32574134229: 3405 ok · 25 newly broken · 7 timeouts · 2 failed · 2 errors of 3441):

• The fix(ci): Cap check-container cores and raise the timeout floor #2875 fixes did what the contention analysis predicted, with one twist. Timeouts went 7 → 5; orbital, SMMAL and VIM now check clean. The five survivors sit at the new 1800 s cap, four of them on both halves — environment-slow, not dev regressions. The twist: _R_CHECK_LIMIT_CORES_ tamed the load tails (shard load p90 6.5–8.0 vs 6.8–11.1 before, peaks 9.0 vs 12.8) but not the medians (~5), and it legitimately slows well-behaved parallel test suites from 4 workers to 2 — which is why PortfolioTesteR and xpect still hit even the raised cap. They, ctmm and E2E are the true long-runners on this hardware; orchaRd timed out on the old half only (new half fine at 1364 s) and is a known flapper.
• The blockCV breakage healed upstream: intSDM and PointedSDMs are back to ok, as predicted.
• The compiler-kill diagnostic fired in production for the first time: gllvm (TMB) failed to install with the manifest saying exactly why — "the compiler was killed inside the container — likely the per-check memory cap (REVDEPX_MEMORY_PER_CHECK)". It had passed twice at 6g with -g0; it is borderline there.
• Newly broken 25 → 26: the stable core of 20 plus the tcltk trio (Boptbd, optbdmaeAT, optrcdmaeAT) all reproduce; c3net and ggm (flappers, broken in 4 of 5 runs now) are in; netrics dropped out (checks clean now); and two packages new to the enumeration enter with real dev-version test failures: nethist and SpaDES.core (its warning-handling test trips over a length-> 1 condition — worth a look on the igraph side).
• The second-opinion machinery ran clean at scale: 783 packages had platform-valid baselines from the previous run (same base image, dependency fingerprints unchanged across 12 days of CRAN churn) — 783 agree, 0 disagree. Old-check reproducibility on the pinned platform is currently perfect.
• hespdiv failed its source-tarball download for the third run in a row — that now looks systematic rather than transient and deserves its own look.

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

🤖 Generated with Claude Code

https://claude.ai/code/session_01UB7YutLzVWU7xCTYUvF3kF