ci: weekly scheduled benchmark run on main — keep the comparison baseline fresh

[aarontrowbridge]

Important

Problem — The published benchmark series (gh-pages, the baseline github-action-benchmark compares PR runs against) only updates when the workflow fires. Gaps between releases leave the anchor stale: as of 2026-08-27 the newest entry was Aug 21, so a PR's benchmark report diffed against six days of drifted main + registry movement — iteration-count deltas that looked like regressions were baseline staleness (Piccolo.jl#317-adjacent campaign, DirectTrajOpt.jl#145 review context).

Approach — Add a weekly schedule: trigger (cron, low-traffic weekend slot, consistent with nightly.yml) so the main-branch baseline never ages more than ~7 days. Auto-push to the published series is already tag/ref-gated correctly for main runs.

Scope — in: one trigger block in benchmark.yml. out: benchmark definitions, thresholds, alert behavior.

Acceptance Criteria

• benchmark.yml carries a weekly schedule trigger
• A dispatched/scheduled run on main publishes a fresh series entry (verified once after merge)