[AIM-IT4]

Summary

Closes #19.

• Add EquityAsianOption.value_mc_sobol(...), using scrambled Sobol Gaussian samples and antithetic equity paths.
• Preserve the existing arithmetic averaging schedule, including before-averaging and partially accrued-average scenarios.
• Use independent Sobol scrambles and return an error estimate from replicate prices rather than applying an IID error formula to low-discrepancy path values.
• Keep all existing pseudo-random MC methods and defaults unchanged. Require num_paths = 2 * num_scrambles * 2**m with m >= 1 to preserve Sobol's balanced random_base2 design.

Tests

Local pytest: 44 passed across test_asian_sobol_qmc.py and the existing test_FinEquityAsianOption.py.

Independent numerical smoke check for a 1-year call (24 annual observations, 20% volatility): randomized Sobol 8,192 paths = 6.59974 (replicate SE 0.02757); existing control-variate MC 100,000 paths = 6.56533 (IID SE 0.00061). This comparison is a sanity check, not a speed/convergence superiority claim.

Branch based directly on upstream master c420937fcec7d9d6382d8cbab0705303db59929c.

CI results pending.

[domokane]

Thanks