MoveFar:mmap dispatches the full chained-probe waypoint vector via
MotionMaster::MoveSplinePath instead of handing a single endpoint
to MoveTo. Removes the motion master's discretion to take a
straight-line shortcut between intermediate points (the diagonal-
through-air bug). Per-waypoint UpdateAllowedPositionZ and vmap
LoS pruning between consecutive waypoints catch geometry-crossing
pairs.
Off-mmap recovery: when the bot's start position has no mmap
polygon (PATHFIND_FARFROMPOLY_START), snap Z via vmap raycast and
NearTeleport so the next tick runs from a valid position.
LastMovement.lastPath populated after every dispatch
(LaunchWalkSpline + new mmap dispatch) so the 10% reuse and
no-worse reuse checks at the top of MoveFarTo can fire — both
gated on a 10y guard so a finished path isn't reused.
MoveFar:spline fallback uses exact_waypoint=false so MoveTo's
SearchForBestPath produces a terrain-following spline.
Per-bot ring buffer of last 3 path attempts on RpgInfo. When 3 mmap or 3 nodetravel attempts to the same dest fail, force the alternative routing strategy on the next tick. When both strategies have failed 3 times each (bothExhausted), fall through to MoveFar:spline rather than flip-flopping forever. Also drops the 10%-per-tick opportunistic combat engage during do-quest travel — the multiplier (0.20x) is the right knob; the random yield was overriding it and producing the 'still grinding too much while traveling' symptom.
Single 75y long-path gate, 40-step chained mmap probe via
WorldPosition::getPathFromPath, spellDistance shortcut to skip the
node graph when mmap can reach the destination, single-waypoint
spline fallback when both fail. MoveWorldObjectTo delegates to
MoveFarTo so quest approaches use the same routing hierarchy.
Removes the stuck-counter teleport recovery from MoveFarTo; the
UnstuckAction layer (planned) is the eventual safety net. Chain
probe fix included: getPathStepFrom uses the explicit-start
CalculatePath overload so each chain step actually advances from
the previous endpoint instead of restarting from the bot's real
position.
