Buck readiness retry exhaustion abandons EPR transition without restoring the standing contract #4

On tulip (RAA489300 3-level buck), enter/exit_epr_mode collapse input current and assert the transition throttle before buck_ready_for_epr(); if its 4 readiness retries exhaust, it reverts the buck mode and returns without sending the CCG EPR command — so no EPR event ever arrives, the event-loss backstop is never scheduled, and the in-progress port bits, 0mA input, learn mode and throttle are left standing. Independent second route into the AC-removal-only stuck state.

Tracking + evidence: tyeth/framework-system#1
Proposed minimal fix (stacked on #3's fix): #2