Scene exits, Back(), and the current/previous scene accessors

A connection between two scenes had to be hand-built as a hotspot: an area,
a cursor, a GoTo, and a flag check written out again for every door. Scene
now carries Exits, and the engine expands each one into a hotspot.

- Exit{To, Label, Side, Area, Needs, Blocked, OnLook, OnTake} in scene.exit.go.
  With no Area an exit is an edge strip, sized as a fraction of Game.SceneArea()
  so a game whose HUD covers the foot of the window gets strips inside the
  painting. Game.ExitLook / Game.ExitTake supply the look and take responses
  once for the whole game, so the phrasing is not repeated per exit.
- Game.SceneHotspots(name) is the scene's own hotspots followed by its exits,
  cached per scene. Authored hotspots come first, so a painted thing beats the
  edge strip wherever the two overlap. HotspotAt and HotspotDebug read it.
- Game.CurrentScene() / PreviousScene(), both persisted in the save file. The
  engine tracked the current scene but exposed no accessor, so every game had
  to shadow it to know where it was.
- Back() returns to the previous scene — what an exit with an empty To binds
  to, for a scene reachable from several rooms whose way out is a direction
  rather than a place.
- Validate rejects an exit naming a scene nobody registered.

Generated hotspots are named ExitName(To) ("exit:street", "exit:back"), so the
location graph can be read straight back out of the registered scenes.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

scene.def.go

@@ -6,6 +6,7 @@ type Scene struct {
 	Background string // Asset.Name
 	Music      string // Asset.Name (optional)
 	Hotspots   []Hotspot
+	Exits      []Exit // connections to other scenes; see scene.exit.go
 	Walkboxes  []Polygon
 	Triggers   []Trigger
 	Actors     []SceneActor