ScipyMinimize(method="TNC") is broken

[terhorst]

Scipy's TNC implementation does not set the number of Jacobian evaluations in the ScipyOptimizeResult, which causes the jaxopt wrapper to throw:

ScipyMinimize(method="TNC", fun=lambda x: x**2).run(1.0)
# Traceback (most recent call last):     
# ...
# KeyError: 'njev'

I'm running jaxopt 0.8 and Scipy 1.11.2.

[mblondel]

I believe this is the same issue as #531.

@zaccharieramzi Looks like some attributes are missing depending on the solver.

[zaccharieramzi]

@mblondel indeed it's probably the same thing.

The scipy doc is a bit elusive on the topic since it says that njev is always accessible normally.

I will try to come up with a quick fix.