fix(ci): Carry revdep2 run ids as strings, and make its summary self-contained

[krlmlr]

The workflow was failing

Run 31084790490
died in plan.R with missing value where TRUE/FALSE needed,
one line after finding its baseline donor:

Baseline donor: run 31048405399 (745 entries)
Warning message:
NAs introduced by coercion to integer range
Error in if (baseline_run > 0) { : missing value where TRUE/FALSE needed

baseline_run <- as.integer(donor).
GitHub's run ids passed .Machine$integer.max in 2026,
so as.integer("31048405399") is a silent NA
that only surfaces at the next if.
This would have hit every run from now on
that finds a baseline to reuse.

Run ids are strings everywhere in these scripts now —
plan.json, the plan job's outputs, the summaries —
with "0" as the "no such run" sentinel
and run_id_chr() / has_run() in util.R to handle them.

Every link in the summary was broken

Checked against the report artifact of
run 31048405399.
Its summary contained exactly 23 links,
all of the form [archeofrag](problems.md#archeofrag) —
right inside the report artifact,
wrong in a job summary served from /actions/runs/<id>,
where they resolve to github.com/igraph/rigraph/actions/runs/problems.md.
Nothing else was linked at all:
the baseline run, the prebuilt donor runs, the retried run and the artifact
were bare numbers or prose.

Now package names link to their CRAN page,
run ids link to their run pages,
and the summary says once
that problems.md and friends live in the revdep2-report artifact.
A regenerated summary has no relative links left.

Packages that could not be checked said nothing useful

All 30 of them appeared as a bare name with version ? and no reason:
revdepcheck's "Failed to check" table is fed a shim
that carries neither.
That section is dropped from the embedded report
and replaced by a table built from the manifest,
which had the detail all along:

Package	Version	Result	Shard	Old	New	Reason
basket	0.10.11	failed	26			old check timed out after 601s
dogesr	0.5.2	depfail	56			Dependencies not installed: rmarkdown, tibble, …
hespdiv	1.2.10	error	101			Source tarball could not be downloaded
optbdmaeAT	1.0.2	failed	38	1E 0W 0N	1E 0W 0N	fails to install against either version

That last row needed a change in shard.R:
an i+ / i- / t+ / t- comparison recorded no message at all,
so three packages had no reason to report.
The shard records it now,
which also makes an i- (installs under CRAN igraph, fails under dev)
readable as distinct from an i+ (broken either way).
The shard summary puts the same reason in the <details> title,
where the log tail cannot cut it off.

I left the failed classification of i- alone.
Arguably that one is newly_broken —
installation regressing under the dev version is a real regression —
but that changes what the reports and the retry selection mean,
so it is worth deciding separately.

How it was verified

No revdep2 run is needed to see the fix work:

• plan.R run against a stubbed gh
  serving run 31048405399's real revdep2-baseline and revdep2-report artifacts.
  Both the baseline-donor path and the retry-run path plan through;
  before this branch, both died on the as.integer().

• collect.R run against that run's real 770-entry manifest,
  with and without unchecked packages.
  The generated summary contains no relative links.

• https://cran.r-project.org/package=<pkg> confirmed as the right link shape
  (200 for a real package, 404 for a bogus one).

• 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_01RLy6bf4Zo8YWgv5Qb53nmY