Fix FX variance swap fair strike and the realised variance normalisation

[Cashubski]

Summary

Follow-up to #282, which fixed the equity variance swap. The FX class has the same replication code and more.

1. FinFXVarianceSwap.fair_strike cannot run. It prices the replication options with FXVanillaOption(self.maturity_dt, k, put_type), the equity constructor's arguments, so every call raises

TypeError: FXVanillaOption.__init__() missing 3 required positional arguments: 'opt_type', 'notional', and 'prem_currency'

The class has no currency pair to build an FX option with, and does not need one: the replication options are now priced with the Garman–Kohlhagen closed form (european_value with the foreign rate as the dividend yield).

2. The same three defects as the equity version. The log-contract drift used r_d instead of r_d - r_f, the option portfolio was compounded by exp((r_d - r_f) t) instead of exp(r_d t), and a put grid truncated to positive strikes was still looped over with the requested count and could contain a zero strike. In FX the foreign rate is never zero in practice, so the first two bias every fair strike by about 2 r_f.

With the fix, under a flat 12% volatility the fair variance strike is within 0.02% of 0.0144 for (r_d, r_f) in {(3%, 0%), (3%, 5%), (1%, 4%), (4%, 4%)}.

3. Realised variance normalisation (both classes). realised_variance sums the N - 1 squared returns of N prices but divides by N. The market-standard definition divides by the number of returns, so the result was low by a factor (N - 1) / N: eleven prices with a constant 1% daily log return gave 0.02291 instead of 252 × 0.01² = 0.0252. Fewer than two prices now raises FinError.

Tests

New tests/unit/test_FinFXVarianceSwap.py (fair strike equals the flat variance across rate pairs; reduced put grid; realised variance of a constant-return series) and one realised-variance test in test_FinEquityVarianceSwap.py. Full unit suite passes (1,065 tests).

[domokane]

Thanks