Fix variance swap fair strike with a dividend yield and reduced put grids

[Cashubski]

Summary

EquityVarianceSwap.fair_strike (the Demeterfi, Derman, Kamal and Zhou replication) has three defects. Under a flat volatility the fair variance strike should be the squared volatility whatever the rate and dividend yield; currently:

r	q	fair variance strike	expected
5%	0%	0.0898	0.09
5%	2%	0.1281	0.09
0%	3%	0.1472	0.09
5%	5%	inf	0.09

1. Dividend yield. The first term used r * t where the log-contract drift is (r - q) * t, and the option portfolio was compounded by g = exp((r - q) t) instead of exp(r t) (DDKZ 1999, eq. 29, with the forward F = S0 exp((r - q) t)). The error is about 2 q, as the table shows.
2. Reduced put grid. When the requested put strikes would go below zero, the code truncates the strike list and sets self.num_put_options, but allocates the weights and runs the pricing loop with the requested count, so it raised IndexError (e.g. 140 puts with unit spacing from a forward near 103).
3. Zero strike. The truncation loop could keep a strike of exactly zero, where the log-contract weight is -inf (the r == q row above).

Changes

• (r - q) in the log-contract term and exp(r t) compounding of the option portfolio.
• Use self.num_put_options / self.num_call_options for the weight arrays and the pricing loops.
• Keep the truncated put strikes strictly positive.

With the fix the fair strike is within 0.2% of the squared volatility for every combination above (the residual is the strike discretisation). The existing test uses q = 0 and is unchanged.

Tests

Two new tests in tests/unit/test_FinEquityVarianceSwap.py: a put grid that exceeds the positive strikes (no error, reduced count, fair strike within 2% of the flat volatility squared), and rate/yield combinations including r == q with a grid reaching zero (finite, within 1%). Full unit suite passes (1,060 tests).

[domokane]

OK. Good find. Thanks.