Show the recent trade window, not just the last print

[piekstra]

The gap

The market rail builds a reference from the orderbook, the trades feed, and the
last print, then objects if the order is materially worse than it. On a thin book
the last print can be days old and far from where the property actually
changes hands — so one stale outlier silently dominates two healthy live quotes.

Observed live: a bid was refused as "6.3% above the market reference", where that
reference was a single print from three days earlier. The orderbook and the
feed both quoted materially higher, and the median of recent prints sat between
them. The rail was arguably right to ask — but it displayed only the number that
made it object, so there was no way to distinguish a stale outlier from a genuine
repricing without leaving the tool and querying the tape by hand.

The change

The diagnostic block now also shows:

• median of the recent window, plus its low/high range
• the age of the last print

  orderbook   bid $63.55  ask $65.00
  trades feed bid $63.66  ask $63.99
  last trade  $59.26, 3d ago   (1890 recent print(s))
  recent      median $61.02  range $59.26–$63.20  (last 20 print(s))

That's enough to tell at a glance that the objection rests on a stale tail rather
than on current pricing.

Trigger semantics are unchanged. This is about giving a human enough context to
judge an override, not about loosening the check — overriding a rail by habit is
how it stops protecting you.

Note on window selection

The window is chosen by timestamp, not array order. The payload is not
guaranteed sorted, so taking the first N would summarise arbitrary prints and
quietly mislead. There's a test pinning this: newest prints appended last must
still be what the median reflects.

Tests

• median over odd / even / empty samples
• describe surfaces the median and range, reproducing the live case above
• the recent window is selected by time, not array order