wall-is-the-layout follow-ups: parked minors from the branch reviews

open   by a73x

From the wall-is-the-layout branch (2026-09-02), all real, none load-bearing:
- wall_layout.persist's doc comment still lists 'a detach' among the saves (exit-time saves were removed).
- pickBirth's redirect (a name the wall already has) also fires for an UNREACHABLE #0 pane with the notice 'Enter there re-creates it' (Enter retries the dial there); CLAUDE.md still calls birthing onto an unreachable row a designed path.
- pickEnd is synchronous on the keyboard thread (bounded by ssh ConnectTimeout=5 + 3 s roundTrip on a dark host).
- /tiles 'state' reads 'connecting' for a healthy pane forever (only gone is ever set); mux.js does not render it.
- hub appendLeaf's anchor fallback (a leaf removed since start) diverges memory order from file order until restart.
- no test spawns a pump to see removePane's detach frame; 'Ctrl-\ x' on a wall of ONE pane untested.
- e2e polish: TIME_SCALE on 06_web's wait done; a layout_leaves lib helper; seed_layout ORIENT for one leaf; rc0 guesses name the first expect; backticks in e2e_10_agent.sh unquoted-heredoc comments; rail_cols counts a notice banner as a rail (fixed in one leg only).
- server.zig stale 'names plus the meta line' comment near stats sizing; SessionEmptied predicate duplicated with server_test_clipboard; two self leaves in a layout are dropped rather than refused as a repeat.