refactor: fold REPL slash commands into one dispatcher

[tgockel]

Repl::run/run_with carried one typed callback pair per slash command (on_tools, on_reset, on_sidecar), so adding /sidecar grew the generic signature by two type params and touched every test call site. The per-command callbacks collapse into a single on_command dispatcher: FnMut(String, Vec) -> Future<Option>, where None means "unknown command". Only /help and /quit stay built in; each caller command's name, help line, and semantics now live together in run.rs (the REPL_COMMANDS table beside the dispatcher match).

/help is composed from structured HelpEntry values: the Repl sandwiches caller entries between its built-in /help and /quit lines and pads one column across all of them. Production output is byte-identical to the deleted HELP_TEXT, locked by a unit test against the old literal. The unknown-command notice echoes the raw post-/ text as typed, and the dispatcher keeps /tools foo (and friends) unknown by guarding on empty args, so observable behavior is unchanged. Two edge quirks of the old exact-match arms are deliberately dropped: trailing-whitespace commands (/quit ) now execute and tab-separated /sidecar args now parse.

Two altitude notes from the 0081 review land in the same pass:

• llm::RebuildingAgent owns the built RigAgent plus its rebuild recipe (resolved agent, cache root, registry), exposing extend_tools / tools / run_turn with the dirty-flag rebuild inside run_turn. This replaces the Rc<RefCell<Rc>> + Rc<Cell> + shared tool-vec plumbing that run.rs threaded through its closures.

• session_setup::SessionRuntime groups the four teardown-bound handles (watcher, mcp_arcs, network, containers; field order mirrors teardown order). teardown now consumes it, and the run/mcp commands assemble it via SessionRuntime::new instead of threading four parallel &mut handles into the REPL state.

Found while designing (filed, not fixed): SIGINT during a turn drops the mem::take'n history before writeback, silently emptying the conversation -- plan/next/repl-interrupt-history-loss.md.

Summary

What does this change, and why?

Checklist

• cargo fmt --all -- --check passes
• cargo clippy --workspace --all-targets -- -D warnings is clean
• cargo test --workspace passes
• Docs in doc/ updated if behavior changed (mdbook build still clean)
• CHANGELOG updated for user-visible changes

Notes for reviewers

Anything worth calling out -- tricky bits, follow-ups, or intentionally out-of-scope items.