Let mock_draft run against the live/projected board, not just past seasons

season_finish() raises for a season with no games played yet, which made
mock_draft unusable for the actual use case of testing the algorithm
against next season's real board -- you could only backtest a completed
past season. Now falls back to scoring picks on the board's own
proj_points when real outcomes don't exist (surfaced as scored_on in the
result and called out in the note), so passing the current season runs
the Monte Carlo bot simulation against this year's live projections
(leak-free through last season) instead.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>