Add mock_draft: Monte Carlo mock draft against simulated ADP bots

Saves the other half of this session's backtest work as a permanent tool.
Unlike draft_backtest, no real draft is needed -- the other teams are bots
that pick by that season's real preseason ADP with reach/fall noise (bigger
swings plausible late, tight consensus at the top) instead of following it
exactly, so who's actually on the board at your turn varies draw to draw.
Your slot (from the active configured league) runs the same recommend()
logic who_should_i_pick uses live, against the same leak-free board
draft_backtest builds.

Runs n_trials (default 30) and reports mean/median/std/range plus, per
round, the most common picks and how often each showed up -- a single mock
draft can make the algorithm look better or worse than its true average
just from bot luck, which is the whole reason to average many instead of
trusting one draw.

Documented in README, docs/tools.md, and a new "validating the algorithm"
section in docs/quickstart.md with concrete example prompts for both
draft_backtest and mock_draft, so they're actually discoverable rather than
just existing.

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