Host multiple workflows in a single sortie daemon (multi-repo use case)

[mihasicehcek]

Problem

Today a sortie daemon is effectively "one process per repository": it watches a
single WORKFLOW.md and all tracker, workspace, agent, and hooks config
lives in its frontmatter. That fits a single-repo user cleanly. It does not fit a
single engineer who owns several repos in the same org:

• Running N daemons means N SQLite stores, N HTTP ports, N dashboards, N log streams.
• max_budget_usd is per-daemon — there is no way to cap spend across all repos a
  user is working on.
• max_concurrent_agents is per-daemon — idle repos can't lend capacity to busy
  ones; headroom has to be partitioned manually.
• Observability and hot-reload that already work inside one daemon have to be
  reassembled externally (systemd / compose / k8s) just to get a fleet view.

Proposal

Let one daemon host multiple workflows, each addressed by a stable name:

sortie --workflow ~/work/billing-api/WORKFLOW.md \
       --workflow ~/work/payments-core/WORKFLOW.md

sortie --workflows ~/work/*/WORKFLOW.md

sortie --config ~/.config/sortie/fleet.yml   # references workflow paths

Existing single-workflow files stay valid unchanged; multi-workflow is opt-in.

What the daemon owns globally

• SQLite store and run/session history (already single-writer WAL — no new contention).
• HTTP server / dashboard, filterable by workflow.
• Optional shared caps: max_concurrent_agents_total, max_budget_usd_total.
• Process supervision and signal handling.

What stays per-workflow

• Tracker config and polling interval.
• workspace.root (kept isolated per workflow to preserve path-containment).
• Agent config, prompt template, hooks.
• Per-workflow caps, bounded by the global caps when both are set.