Grade trail difficulty on ascent, not distance alone

[glennmichael123]

Part of #1002. Blocked on #1003.

Problem

Trail difficulty is inferred from distance alone, because ascent is not available. Across a 400-trail production sample that produces 310 easy, 21 moderate, 69 hard — a grading that says almost nothing true:

• A 0.1-mile scramble straight up a cliff grades easy.
• A flat 250-mile rail-trail grades hard.
• /trails sorted longest-first therefore opens on a wall of national scenic trails all badged HARD, which is the first impression the catalog makes.

Every difficulty filter in the product is currently sorting by distance a second time, and the badge on each card is telling people something we have not measured.

Acceptance

• Difficulty derived from distance and ascent once Backfill elevation gain for the whole trail catalog (0 of 400 sampled have it) #1003 has filled it.
• A trail with no elevation data shows no grade rather than a distance-derived guess — an absent badge is better than a wrong one.
• The rule lives in one shared pure function used by the API, the cards and the filter, so the three cannot drift.
• Reviewer difficulty votes (already collected) stay separate and keep being shown as what people who walked it thought.
• Tests covering the short-and-steep and long-and-flat cases that are wrong today.