sylva-units Helm chart rendering memory consumption is high

We've known for a while that the memory consumed by Flux helm-controller is high during rendering of sylva-units Helm chart.

We've regularly increased the memory limit of that pod to compensate for that, but this isn't sufficient:

• without also increasing requests it does not influence pod scheduling, and the scheduler may place the pod on an otherwise loaded Node, creating conditions for memory pressure conditions on that node
• increasing requests for something which isn't a frequent leads to pessimistic dimentioning, resulting in less efficient resource usage

We are aware of some occurrences, in particular in CI, where this creates breakage (e.g. #3425).

With this in mind, we would benefit from improving the memory footprint of Helm during processing of sylva-units Helm chart.

(Related to sylva-projects#102)