lmer: poly of time as fixed effect but factorized time as random slope?

In my mixed-effects model (lmer from lme4) I have a four-way interaction as a fixed effect including

1. a cubic polynomial of time (training sessions)
2. Age
3. Food dosis (half of the animals having 0 gr, whereas the rest varying from 1 until 7 gr)
4. Condition (4 levels).

My question concerns the issue that a polynomial as random slope has been discussed as over-complexifying the model and my model also does not converge when using poly in my random effects. When plotting the interaction and dichotomizing food and age for visualisation of my data, I see that for the animals with food, the polynomial course is apparent across time for all conditions in a similar pattern, whereas no food intake was associated with a linear trend.

Two ideas about alternatives of using poly as a random slope:

alternative 1: just use time as a linear effect as a random slope, while keeping the polynomial time effect as a main effect

alternative 2: factorize time into lets say weeks (3 weeks in total), because the polynomial effect for the animals with food followed a weekly pattern.

Comment on 2: of course this argumentation is somewhat arbitrary, therefore I was wondering if I could get some advice on a better solution.

Thank you so much for your assistance!