Multiplexer: add Herdr 0.8 alongside tmux

[skeletor-js]

Method: CLI (multiplexer)

I would like to add Herdr 0.8.x protocol 19 as an additive terminal backend while keeping tmux as the default and preserving existing tmux registry identity.

Pane identity

A process inside Herdr receives explicit terminal/session/socket context. The stable Harness identity would be the configured endpoint/session namespace plus Herdr's terminal_id; pane_id remains a mutable route because terminals can move between workspaces. This keeps different Herdr endpoints and tmux panes collision-free.

Proposed shape

• Keep tmux as the default and support TERMINAL_BACKENDS=tmux, herdr, or tmux,herdr.
• Put both implementations behind one backend-neutral contract for inventory, PID/cwd discovery, literal text, logical keys, capture, notifications, titles, and lifecycle operations.
• Use Herdr's versioned local NDJSON socket API for production control rather than passing prompt text through CLI argv.
• Carry namespaced terminal identity through discovery, registry persistence, hooks, reconciliation, deletion, and recap-environment scrubbing.
• Migrate legacy tmux registry data additively so existing agents are not orphaned.
• Treat backend outages as unknown/unavailable rather than false process death, and never automatically replay an input operation after an ambiguous transport failure.

I have a working implementation prepared for review. Its automated real-backend suites pass against tmux and Herdr 0.8.0 in managed Linux environments, including lifecycle, notification, input, capture, identity, workspace movement, outage handling, nested multiplexer topology, and process discovery.

Validation caveats

• I have not yet performed hands-on validation on the intended physical device because it has not arrived.
• All 12 installed agent binaries passed process discovery on both tmux and Herdr in the managed test environment.
• Authenticated model turns were not exercised for providers where credentials or verified no-incremental-cost entitlement were unavailable. Those rows are reported as discovery-verified, not inference-verified.

Does this identity and additive-backend shape match the direction maintainers want for a Herdr contribution?