jnp.interp precludes reheating scenarios

[cgiovanetti]

jnp.interp expects monotonic increasing input. abundances, weak_rates etc interpolate on photon temperature, so this breaks for reheating scenarios. Possible solution is to use interpax.interp1d in everything except thermo (where we're interpolating pretabulated results that we know are monotonic); performance needs to be tested

[cgiovanetti]

Looks like two jnp.interps were missed in weak_rates, need to check if there are others missed as well

[cgiovanetti]

And more introduced by 76101eb

[cgiovanetti]

I am also finding that some modified versions are running much slower (>10s), flagging this to look into