Add qb_boost to model_settings: a testable QB valuation tilt

New lever, distinct from every existing weight. oline/pace/schedule/etc
all scale a real per-player signal; qb_boost is a direct fractional lift
on QB draft_score you supply because you believe the position is worth
more than the projection says, not because of any one player's inputs.

Comes from a chain of checks this session: champion_strategies initially
suggested mid-round QB was a winning pattern, but that turned out to be
survivorship bias from 6 champions. Testing the full real draft history
of one league instead (684 picks, 2020-2025, via draft_backtest's value
methodology found something more specific and better supported: QB beat
its actual draft cost in that league in every one of 5 measurable
seasons, by 20 to 55 spots on average, an 74 to 95 percent beat rate each
year. Not one hot year -- five straight. Consistent with the known
broader dynamic that QB scoring has outpaced ADP since only one roster
slot is required regardless of how much the position scores.

Applied as a straight multiplier on QB draft_score in project(), after
consistency blending, before overall_rank -- verified it moves ranking
(a 15 percent boost took Josh Allen from rank 4 to rank 1 in a test
league) without breaking roster-need decay (a second QB still isn't
recommended once you have one; the boost affects how good QB1 looks, not
whether you need a QB2). Defaults to 0.0, no change to existing behavior.

Documented as league-specific, not universal -- the docstring and
docs/tools.md both say to verify the underlying pattern for your own
league via champion_strategies/draft_backtest before setting this above
0, since it is a real, tested finding for one league, not a general
claim about quarterbacks.

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