Can HPI be accelerated in opt savings and opt investment?

[jstac]

Right now, in the final figure, OPI is beating HPI for some values of m in the final figure of opt_savings_2 and opt_invest.

Can HPI beat OPI if we loosen the tolerance or change other parameters in https://jax.readthedocs.io/en/latest/_autosummary/jax.scipy.sparse.linalg.bicgstab.html ?

Note that #165 should be done first, so that we have a visual inspection to show that we don't loose too much precision when we make these changes.