Marks past the wall's ends are still listed, exported and measured #42

Seen in device field test run 2: TestFlight build 3.1, beta at ad91620, capture code t3/ios-mvf e9e405d, server e106acb. The exported scan was lost before it was copied off the phone, so this comes from the screen recording and the code at ad91620. Times are in the recording. b7f51ba and 49a676e on t3/ios-mvf don't change this.

What happens. The final wall ran from 7 ft 6 in left of the meter to 0 ft 6 in right (see #24 and #35). The feature
list still showed "AC unit / About 16 ft right of your meter" and "Window / About 7 ft right of your meter". The result
measured both:

• doors and windows: "Measured 5 ft 11 in. The rule is 3 ft…";
• AC: "Measured 15 ft 6 in…".

Run 1 showed the same with the AC stand-in 16 ft past the right end.

Cause.

• App: the review lists every feature (MarkFeaturesScreen.swift#L106-L110), and the
  export writes every object with no check against the wall's extent (ScanEngine+Export.swift#L30-L51,
  SceneExport.swift#L396-L420).
• Server: parse_scene only checks the wall_id (scene.py#L760-L769). A window without a footprint
  becomes a line on the wall's straight extension past the end (scene.py#L713-L740), even when that end
  is unexplored and may be a corner.

This is harmless in this run, because the marks were far from the spot. But it tells the homeowner their marks count
when the scan no longer covers them. It also puts objects on a wall line that may not exist.

Possible fixes (your call):

• in the app: flag marks outside the ends on the feature list ("past the end of your scan"), or extend the end to
  include them;
• on the server: say which objects lie past an unexplored end, and don't treat them as on the wall's line.