Per conversation starting around here, many of the "good" techniques for resolving pose ambiguity rely on doing some work on the RIO side (multiple camera fusion, wheel odometry, etc.)
Add support to get all possible target poses through the libapriltag library, through JNI, through detection results, through NT and photonlib.