Precision issues

[DillonJ]

There seem to be precision with SpineInterface. For example, we have this:

Yet we get this:

Window 2: 2030-02-28T00:00~>2030-03-01T00:00
CPXPARAM_MIP_Tolerances_MIPGap 0.050000000000000003

We have had this issue cause infeasibilities, with small values etc. and when zeroes aren't zeroes but very small floats.

It would be good to understand where this imprecision is coming from?

I'm thinking perhaps we should be able to specify a maximum precision in our call to using_spinedb(). This would have many benefits for the underlying optimisation problem, where small coefficients, for example, can be replaced with zeroes.

Any thoughts @manuelma ?