Team Bus: let an agent add/edit its own session's run commands, not just title/color/status

[AThraen]

Feature request

Team Bus's self_set MCP tool (see docs/team-bus/spec.md §3, implemented in src/TeamBus.Service/Mcp/SelfTools.cs) already lets an agent control its own shell's display: title, status, color, focus, notify. It should be extended (or a sibling tool added) so an agent can also add/edit its own session's run commands — the same RunCommandItem list a human operator manages today via the sidebar ▶/▼ menu and SessionRunCommandsDialog (see Models/RunCommandItem.cs, Views/SessionRunCommandsDialog.xaml(.cs), and CLAUDE.md's "Per-Session Run Commands" section).

Why

An agent working autonomously in a session often knows the right build/test/lint command for the project it's in (or discovers one over the course of a task) before a human operator has configured it. Letting the agent register that command as a labelled run command — the same object a human would create through the dialog — means:

• The ▶ button / F5 keybinding becomes useful without the operator having to pre-configure anything.
• Other teammates (or the operator) can re-run the same command later without re-deriving it.
• It's consistent with the existing "agent can control its own shell's presentation" precedent self_set already established — this extends that precedent from display to this other per-session configuration surface.

Scope notes (not designed yet — for whoever picks this up)

• Should stay scoped to the calling agent's own session, exactly like every other self_set-adjacent capability (no way to target another agent's shell).
• Needs the same schema-conservative design discipline as the rest of the MCP tool surface (spec §3's "keep tool schemas flat, no oneOf/anyOf, assume the least capable MCP client").
• RunCommandItem has Mode (Process/PowerShell) and PostRunUrl fields — PostRunUrl in particular auto-launches a URL on successful exit with no confirmation (see CLAUDE.md's warning about RunInstance.IsLaunchableUrl), so an agent-supplied value here needs the same validation/trust treatment as the existing human-supplied path, not a looser one just because it's new.
• Whether this becomes a new tool (self_add_run_command/self_edit_run_command) or an extension of self_set's existing shape needs a real design pass — not decided here, just filing the idea.

Requested by the maintainer during team-bus feature development; not yet scoped into docs/team-bus/plan.md.