fix: tmux missing from devShell — the real cause of the tmux test flake
The actual root cause of test_e2e_tmux_secondary_da_no_corruption's
"flaky" timeouts under real ci-tron runs: tmux isn't declared in
flake.nix's devShell (unlike sway/grim/wlrctl/wl-clipboard, all added
for the exact same reason — see the previous two fix commits), and
ci-tron.service deliberately gives its runner a curated, minimal PATH
with no /run/current-system/sw/bin, so nothing puts tmux within reach
there. `tmux -f /dev/null new-session /bin/sh` fails outright; since
PS1 is cleared, the outer shell's own prompt is invisible, so the
test's "$ " marker (the nested shell tmux was supposed to start)
simply never appears — e2e_pump_until times out at whatever bound it's
given, not because anything was slow.
Traced by watching the process tree of a real, isolated ci-tron test
run at 1.5s resolution: tmux never appears at all, at any point in the
run's lifetime — ruling out the timing-race explanation the previous
commit assumed. Confirmed by reproducing (deterministically, 5/5) via
systemd-run configured to match ci-tron.service's exact hardening
properties *and* its literal PATH value read from the live unit, then
watching the same reproduction turn green once tmux is added here.
See the corrected comment on e2e_pump_until in e2e_test.odin for the
same account in-place, so it doesn't keep pointing at CPU contention
the next time someone reads it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KpDBFeBQKx9f5U3pfVeVV3