Dispatch routing hardening, new /dispatch page, Hub UI polish.

Added MiniBrain pack base tiers, tier-aware generation hooks, and mb-base mirror tags while enforcing private Docker Hub publishing for pack images.

Updated QA workflow defaults to build all tiers and added manual `tier_group` selection for `all`, `openclaw`, or `pack` runs.

Swapped non-fast-forward retry sync from rebase to fetch+merge and tightened follow-up prompts so failing log paths and root causes are reviewed first.

Upgraded Hub UI behavior with emoji-mart picker parity, refreshed action copy from Queue Task to Run, and adjusted prompt pane spacing and copy.

Closed merged PR tasks automatically in monitor cleanup, removed enforced `moltenhub-` PR-title prefixing, and refreshed README run-command docs.

Hardened hub connectivity loops with transport-aware presence sync, websocket reconnect handling, HTTP fallback gating, and unauthorized-error exits.

Added schema-aware manual-dispatch placeholders and payload hints so selected skill definitions surface expected input shape in Dispatch UI.

Simplified README onboarding with direct molten.bot/dispatch docs link plus quick Docker run command examples.

Launched and iterated the `/dispatch` landing page with richer product positioning, structured data, runtime walkthroughs, and Docker command copy flow.

Refreshed `/code` and supporting site pages with new layout/content passes, updated dispatch screenshot assets, and footer navigation ordering updates.

Updated task failure responses to include explicit `Error details:` messaging plus structured `error_details`, and tightened home-page agent copy.