fix: honor selected dashboard review capacity

[cr-shah]

Finding

The Review queue slider showed the nearest saved capacity results and kept the transaction queue at the generated 750-review ranking. A selection such as 75 therefore displayed results for another capacity.

Fix

Calculate all three policy scorecards and the expected-loss queue from the scored test transactions at the selected capacity. Keep the generated artifacts unchanged.

Verification

• Added a regression test for 75 reviews and queue ranks.
• Full pytest suite: 58 passed.
• Ruff: passed on tracked Python files and changed files.
• Offline lock check and installed-package compatibility check: passed.
• Fresh 12,000-row demo outputs matched checked-in figures and data byte for byte.