fix(curiocity): fail all-skipped runs as no runnable trials

[atomicdjt]

Fixes #326.
Summary
Treat a non-empty Curiocity run in which every resulting trial is skipped as having no runnable trials, returning ExitCode.CONFIG_ERROR (2) instead of success.
This preserves the existing semantics for individual skipped trials: a skipped trial remains non-fatal when at least one other trial is runnable.
Changes
return exit code 2 when every selected trial resolves to skipped
preserve existing mixed runnable + skipped behavior
add unit coverage for both cases
add an end-to-end regression covering the missing/mistyped-agent path from #326
Verification
Verified on commit b80f055:
CI Curiocity ✅
type validation ✅
package build ✅
full Curiocity test suite ✅
CodeQL Advanced ✅
The implementation intentionally reuses the existing CONFIG_ERROR / exit-code-2 contract for “no runnable trials” rather than changing per-trial skipped semantics or introducing a new exit code.

[github-actions]

Rosetta Triage Review

Summary: Fixes #326 — a Curiocity suite where every selected trial resolves to skipped (e.g. all agent ids mistyped) now returns ExitCode.CONFIG_ERROR (2) from gatekeeper() instead of 0, matching the documented contract in docs/architecture.md §13. Mixed runnable+skipped suites keep prior behavior (a lone skipped trial stays non-fatal). Also adds a repository == 'griddynamics/rosetta' guard to repo-implement.yml/repo-plan.yml so those scheduled/issue-triggered automation jobs don't fire on forks.

Findings:

• Root cause and fix match the issue's own diagnosis: run.ts:244 guarded on matrix.length === 0 (never true when agent ids are merely unresolvable), not on the built trial specs; the new check in gatekeeper() (trials.length > 0 && trials.every(status === 'skipped')) closes that gap at the aggregation layer, which is shared by both run and report (D8), so a re-gated old run also gets the corrected exit code.
• Test coverage is solid: a unit case for all-skipped, a unit case confirming one skipped + one runnable trial still exits 0, and an integration test reproducing the exact [ROSETTA] curiocity exits 0 when a run executes zero trials, so CI can pass on nothing #326 scenario (mistyped agent → 1 skipped trial → exit 2).
• The fork-guard changes to repo-implement.yml/repo-plan.yml are unrelated to the stated PR scope (curiocity gatekeeper fix) but are low-risk, defensive additions consistent with existing repo-scoping patterns elsewhere (this repo's automation is board-driven and org-scoped already).

Caveats:

• Scope mixing: the PR bundles an unrelated CI/workflow change (fork-guard on two automation workflows) with the curiocity bug fix. Consider splitting in the future for cleaner history/review, though not blocking here given the small, low-risk diff.

Automated triage by Rosetta agent