Share one control loop between straight() and turn(), compensate for battery voltage

[SaintSampo]

Summary

straight() and turn() were near-duplicate loops solving the same problem on different axes. Both now delegate to a shared _move() that controls translation and rotation together.

_move() runs translation and rotation through their own controllers and mixes them as left = translation - rotation, right = translation + rotation. straight() passes heading_target = 0, turn() passes distance_target = 0, so each move drives one axis and actively holds the other at zero. A zero goal doesn't gate the exit, so both functions still exit on the same axis they did before.

Changes

• Effort limits apply to the mixed wheel efforts, not the main PID's raw output. Both wheels scale by a common factor, so heading correction survives saturation instead of being clipped. min_effort is a new optional argument on both functions.
• The floor only engages while an axis is outside tolerance. Otherwise a settled axis lets a tiny correction on the other get scaled up to min_effort, spinning the robot in place over a fraction of a degree.
• Battery voltage compensation. Effort is raw PWM duty, so a gain set tuned on a fresh pack is 30-40% weaker near cutoff. __init__ averages 8 readings into voltage_scale, applied in _move. max_effort now means fraction of nominal power, so duty can exceed it on a weak pack. arcade() unchanged.
• Retuned both boards. Integral terms dropped on the XRP; tolerance_count 3 → 10 on both, since at the measured 11.2 ms loop period 3 was a 34 ms dwell — short enough to latch "done" while still moving through the band.

	XRP	Nano
min_effort	0.14	0.10
distance kp / kd	3.5 / 0.1	0.32 / 0.0184
distance tolerance / count	0.1 cm / 10	0.2 cm / 10
heading kp / kd	0.064 / 0.0045	0.014 / 0.001
heading tolerance / count	0.5° / 10	1° / 10