Follow-ups left open by the station client (#464)

[josephnef]

Left open by #464 (merged as 927cfc7): the station client over IRadio + src/sta, its headless cells and tests/mt7612u_sta_onair.sh.

From the review

• Back-to-back hostapd restarts race the AP interface. cell_end returns as soon as the hostapd process exits and ap_up launches the next one at once. One restart in four on the review rig started 30 ms after the previous AP-DISABLED and died with Could not read interface wlp13s0u2 flags: No such device / nl80211 driver initialization failed (scored INCONCLUSIVE, correctly). Wait for the netdev to be present and type managed before launching, or retry once on that failure. Fits the sibling-harness cleanup PR the author mentioned (INT trap not exiting in mt7612u_ap_onair.sh, hostapd left running when mt7612u_sta_identity.sh is interrupted during AP start, hand-back while the process is alive, unbounded waits).

From the PR description

• Realtek DUTs in the on-air cell. The client arms an 8822C / 8822B through the jaguar1/2/3: the Realtek station arm for IRadio::SetStationIdentity #463 arm, but sta_dut_take is MT7612U-only; a generic DUT take / hand-back is needed before the cell can cover them.
• A reconnect cell (AP stopped and restarted): the re-join policy is pinned headlessly only.
• Software-CCMP throughput under load, and a long soak with the default retry settings.

Unscored observations

The noarm and retry0 INFO lines (link outcome, not scored) read 0 % loss on most runs, 33 % and 16.7 % once each. Six pings at near field is a sample; if these cells ever score the link, they need a real measurement window first.