I think what you are asking here is a complex import mapping setup. I don't understand the full scenario, but here are some possibilities:
> In order to be able to query or write individual values, the paths of the existing assets must first be read.
This sounds like solve-able by using "find by key" approach in import mapping. You can also use "find by microflow" where you can handle finding the path on your own. This is the most powerful way available in import mapping.
> They are listed as array elements, whereby new assets can be added at the end and those that are deleted simply disappear from the array.
The "find by key" approach would create object by default when the key is not found.
To delete an object during import mapping is not possible. You could introduce timestamp to indicate when was the last time this object being updated. With this information, you can clean the objects afterward.
> on the one hand, there is no fixed JSON structure that can be expected from a get call
This contradicts with the earlier statement, `which are always structured according to the same standard`. In the case "structure is not fixed but predictable", you can simply receive it as a string in microflow, then handle the import mapping as the next activity depending on the type of payload. However, if the structure is unpredictable, the only option is to handle it yourself.
Hopefully this information helps.