Review fixes, no logic change:
- T17a now asserts the restore is the first write once the charger is back
(it was only checked to happen at some point); mutant M3 (restore after
SETTLE_S) now turns T17a red.
- T15 now asserts the refused restore is counted once (`fails == 1`);
a mutant without `_failed` there now turns T15 red.
- DECISIONS "The phase sequence" and the carmode docstring reflowed to 79
columns; DECISIONS says D counts as drawing too, as the code does.
Co-Authored-By: Claude <noreply@anthropic.com>
Stage zero counted mode3 B as quiet whatever the power and selected at the
20 s timeout under load. A power reading now decides (under 100 W); mode3
(not C, not D) only without one. Still drawing at 20 s: no select, the
current comes back at once, one failed try, next try PHASE_GAP_S later.
Co-Authored-By: Claude <noreply@anthropic.com>