Fix off-by-one in the Bachelier implied-volatility denominator polynomial

[Abhayindia]

put_iv panicked with index out of bounds: the len is 9 but the index is 9 because the denominator loop ran 1..=B.len(). The clippy pass in c9ae943 rewrote that to .skip(1), which stopped the panic but left the two inverters summing different polynomials - and both were wrong anyway. The Choi-Kim-Kwak denominator is 1 + sum_{k=1..9} b_k eta^k, so B[i] carries power i+1, not i. It's invisible at the money (eta is small, the denominator barely matters) and gets bad fast on the wings.

Round-tripping price -> iv through the model's own pricer, f=100, t=1. Before:

       k      sigma        call_iv         put_iv     call_err      put_err
    90.0       5.00     1.81823845     1.94645854      6.36e-1      6.11e-1
    90.0      10.00     6.68362009     6.72652954      3.32e-1      3.27e-1
    95.0      10.00     8.85693656     8.87351931      1.14e-1      1.13e-1
    99.0      20.00    19.97385861    19.99557600      1.31e-3      2.21e-4

After:

       k      sigma        call_iv         put_iv     call_err      put_err
    90.0       5.00     5.00000000     5.00000000     6.29e-12     6.29e-12
    90.0      10.00    10.00000000    10.00000000     3.02e-13     3.02e-13
    95.0      10.00    10.00000000    10.00000000     1.24e-15     1.24e-15
    99.0      20.00    20.00000000    20.00000000     8.88e-16     1.07e-15

Added a round-trip test over strikes and vols - it fails on master and passes with this change. The 25 pre-existing failures in finite_difference_pricer / longstaff_schwartz / sabr (#272) are untouched: 62 passed / 25 failed both before and after.

Closes #312.