[arellano] Default occurs at zero debt due to incorrect `B0_idx`

Lecture: Default Risk and Income Fluctuations (lectures/arellano.md)

In ArellanoEconomy.__init__:

self.B0_idx = np.searchsorted(self.B_grid, 1e-10)

With the default grid, B_grid[125] is exactly 0.0, but searchsorted returns the first index >= 1e-10, i.e. 126, where B_grid[126] = +0.0036.

This breaks two things:

1. Default at zero debt. T_d uses v_c[B0_idx, :] as the re-entry
   continuation value, so default carries a small reward. Since h(y) = y
   for y < 0.978, default is costless at low income, and this is enough
   to make v_c(0, y) < v_d(y) there — even though h(y) <= y and
   v >= v_d imply the reverse must always hold.

2. Positive assets while in default. simulate sets Bp_idx = B0_idx, so foreign assets sit at +0.0036 throughout every exclusion period.

Proposed fix

# The index of the grid point closest to zero
self.B0_idx = np.argmin(np.abs(self.B_grid))
self.B_grid[self.B0_idx] = 0

Selecting by distance is robust to floating-point noise; the second line also covers grids that do not contain zero (e.g. B_grid_size=250), and is a no-op with the default grid.

Note that this changes the bond price schedule and the time-series
figures, so they will need regenerating.

Happy to open a PR.

Start in lectures/arellano.md, reading ArellanoEconomy.init and simulate around B0_idx, then inspect how T_d uses the re-entry value. Verify that zero is selected for the default grid and for grids without an exact zero, default-state assets remain zero, the value inequality is restored, and the bond price schedule and time-series figures are regenerated.