BDSP sim: measured walk/run pace and map-crossing stall, independent of the walker's table #217

Part of #139. Related: #203.

The walker counts tiles from hold time, and GridSim (#167) replays holds with route_ms_per_tile(), the walker's own pace table. A wrong pace therefore moves the walker and the sim the same way, and no sim test can fail on it. GridSim also has no pause when a walk crosses into the next map.

FRLG hit this on 2026-11-03: a 20-tile running leg from Cerulean City onto Route 24 ended one tile short twice, and the log claimed the planned tile. The walker's open-leg margin is about 67 ms running, so either a map-load stall of that size or a pace a few ms per tile slow explains it. BDSP has the same blind spot.

• Measure walk and run pace per direction on console B from the Stream History video, not the computed [Speed] line: several legs to a wall each way.
• Measure the stall when a held walk crosses a map connection (a route-to-town edge, e.g. Route 201 to Sandgem).
• Give GridSim its own measured pace table and the connection stall, separate from route_ms_per_tile().
• A test that fails first if the walker's pace or margin doesn't cover the measured numbers, from a real crossing leg.
• Fix the walker (pace table, margin or a crossing allowance) and confirm on the console.

FRLG's overworld sim is tile-only and gets the same model; that is tracked with the FRLG story work.