flutter_scene_net 0.2.0

• PredictedTransformComponent with PredictedController, client-side prediction of an owned entity with authoritative input-replay reconciliation, so local input renders instantly while the sim stays server-authoritative. Selected per entity through SceneReplication's localPrediction seam.
• PredictedPhysicsComponent with PredictedPhysicsController, rollback prediction for an owned physics body. Mispredictions restore a retained world snapshot, adopt the authoritative pose, and replay pending inputs through the simulation. Needs a snapshot-capable backend (rapier).
• PredictedPhysicsController.onWorldRestored fires after a rollback restore. A restore rewinds to the body set its snapshot captured, so a controller managing bodies dynamically (remote-player proxies) allocates a fixed pool before the first snapshot rather than recreating them.
• NetworkTransformComponent re-anchors its interpolation buffer after an idle gap, so a resuming remote eases in instead of snapping.
• The interpolation delay adapts to measured arrival cadence and jitter, easing between minDelay and the configured ceiling, so clean links render remotes barely behind while jittery ones buffer deep.
• SceneReplication.inputTargetDepth passes the input buffer depth through to the client (1 suits local or stable links).
• SceneReplication.minInterpolationDelay and SceneReplication.adaptiveInterpolation reach the interpolation buffer, so an app can floor the adaptive delay or hold it fixed.
• Both predicted components reconcile at the tick the authoritative pose was snapshotted at, not the input ack's, so a deferred snapshot no longer discards the travel in between. They wait for a snapshot to date the pose before reconciling at all.
• Both predicted components take maxCatchUpTicks and snap to authority past it, so a backgrounded tab or a long hitch does not replay every missed tick in one frame.
• Requires dashwire_replication 0.2.1 for the input-command and prediction substrate.