fix(navigation): a place in a dungeon cell stands on the floor under it

[davidc-smith]

What this changes

A walk or preview to a place inside a dungeon kept the height it was given. A dungeon map gives a cell by the middle of its bounds (PluginDungeonCell.Center), which in a tall room is metres above the floor. The router only arrives within 1.2 m of a goal's height (GoalHeightTolerance), so such a place came back NoRoute. Even a 3 m room's middle is 1.5 m up. A plugin exploring dungeon 0x5B48 cell by cell got NoRoute for every one of its ~700 cells; in 0x5948 only some routed. Walks to creatures were unaffected, since a creature already stands on the floor.

RuntimeNavigationGoalSource.TryLocatePlace now stands a place in an interior cell on the floor under it:

• the cell's own floor nearest at or below the point, allowing up to 0.5 m above it so a place given at floor height stays on that floor;
• where the cell has no floor under the point, as in the upper half of a tall room authored as a cell on top of a cell, the nearest floor of any cell below it in the landblock.

Both the preview (PreviewPathAsync(position, …)) and the walk (GoTo(position, …) / WalkToPlace) locate places through TryLocatePlace, so they agree. Outdoor places, places given with a NaN height, and object goals are unchanged. CellSurface.SampleLevelZAtOrBelow is a new lookup that only considers level faces at or below the point, so a ceiling over the point is never picked; SampleFloorZ returns whichever face it meets first.

No plugin contract change: the abstractions only gain XML doc notes that a cell's Center can be passed as it is. docs/navigation.md is updated to match.

Tests:

• DungeonPlaceFloorTests (made-up dungeon with a 3 m room, a 5 m room, and a 10 m room built as a floorless cell over a floored one):

  • a goal left in mid-air returns NoRoute today;
  • a tall cell's middle routes and the walk arrives;
  • a floorless upper cell routes to the floor of the cell below;
  • a low room, outdoor places and places with no floor under them are unchanged;
  • a place a little under its floor stays on it.

  With the fix stubbed out, 5 of the 10 fail.

• CellSurfaceTests: the new lookup takes the floor under a point, not the ceiling over it.

• DungeonPlaceFloorInstalledDatTests (InstalledDat lane, Diagnostic): routes to the middle of every cell of 0x5948 and 0x5B48 from one start, as given and stood on the floor. Results:

  Dungeon	Routed as given	Routed on the floor
  0x5948	39 / 327	313 / 327
  0x5B48	9 / 712	265 / 712

  The rest of 0x5B48 lies in wings its cells alone don't connect.

It has also been tested live in a client against both dungeons.

Original behavior

Not a change to how the client behaves in the world. It only affects where the client's own navigation layer puts a place goal before planning a route to it.

Checklist

• dotnet build AcDream.slnx -c Release is green
• The portable test filter passes locally (see CONTRIBUTING.md)
• One topic per PR
• No code copied from ACEmulator, ACViewer, or holtburger
• No comments describing the original client's internals

https://claude.ai/code/session_01CJ7QWFYQz9G3sBYnBWWDzU