Contract 6.15.0

The Elixir Gym's second war run (feedback #84-#86), on the day-by-day and the exact-week path of war_history.

• days[].standings[].progress_end_banked (and on war_current.days_closed[]): the race log caps a finished boat's progressEndOfDay at the line, so the finishing day read 10000 where 6811 + 3000 + 323 = 10134 was banked and the next day's progress_start carried it - the one row in twenty where the array's own arithmetic broke, and a walk over progress_end showed +134 on a day that earned nothing. progress_end stays the API's value; progress_end_banked is the banked one on every row, equal to it wherever no cap fired; a conditional note names the clamped rows.
• A conditional note on war_history.member_weeks[] and war_current.participants[] whenever a row carries boat_attacks > 0: boat attacks are counted INSIDE decks_used and scoring_decks, a boat battle spends a war deck and scores on a different scale, so points / scoring_decks is not comparable between those rows and the rest (the rate the 6.11.0 note sanctions pooled the two: a member with four boat decks ranked last of 26 on it). The decks note now lists a boat battle among what consumes a deck; the docs do too.
• history_starts_at on the exact-week path as well (it was served only with seasons), and an empty exact week says which side of the horizon it is on: before recording began, after the latest recorded week, a section no season has (0-4), or a gap inside the span - a week before the horizon and a week that never existed answered byte-identical.

Additive.