t27f: list every spec on an import cycle (deps_cycles in spec_deps.t27) #10693

The T27F dependencies view (specs/t27f/spec_deps.t27, #10680) flags only whether the focus spec is on a cycle. Its own honest-limits block says: "A cycle among other specs is not reported." #10682 was found only because a corpus run asked the question for every spec, one at a time, in a scratch loop. After #10689 the corpus has no import cycle. But the view has no law that answers "which specs are on a cycle?" for the whole edge list, so a new cycle has nothing to check it against.

Competitors make this check a rule, not a view. dependency-cruiser has a no-circular rule and madge has --circular: both list every module on a cycle and fail the build. Bazel and Nx refuse a cyclic target graph outright.

Boundary

• In scope: specs/t27f/spec_deps.t27 (a new function, its tests and the header's honest limits), its seal, and the SKILL.md row.
• Out of scope: a host that runs it on every PR. That needs t27c run (t27c: no subcommand runs a spec's main, so T27F views have no host #10692), or a CI job, and CI jobs are gate files.

User Scenarios

• Given the use-edges list of any tree, a reader gets every spec that is on an import cycle, each once, and an empty list for an acyclic tree.

Requirements

• deps_cycles reuses deps_walk and deps_on_cycle. It copies no body (duplicate-bodies).
• Each importer is walked once. Specs that import nothing are never on a cycle and are not walked.
• If the output, a walk's set or the walked list does not fit, or a walk goes deeper than DEPS_MAX_DEPTH, the answer is refused whole (cap + 1), never cut off.
• Tests must cover:
  • the old igla pair, where both specs are listed;
  • a DAG, which gives an empty list;
  • a cycle the first spec is not on;
  • a self-import;
  • two separate cycles plus a repeated line;
  • the refusals.

Success Criteria

• t27c test-report specs/t27f/spec_deps.t27 passes with 0 vacuous tests on the Railway lab.
• On the lab, a scratch test that embeds master's use-edges list (not committed) finds 0 specs on a cycle after igla race: break the backend <-> rtl import cycle (Closes #10682) #10689. On the list from before igla race: break the backend <-> rtl import cycle (Closes #10682) #10689, it finds exactly specs/igla/race/backend.t27 and specs/igla/race/rtl.t27.
• python3 tools/dupe_scan.py is ok on the lab.

Success Criteria

• t27c test-report specs/t27f/spec_deps.t27 exits 0
• t27c test-report specs/t27f/spec_deps.t27 reports 0 vacuous tests
• python3 tools/dupe_scan.py exits 0