LLM-powered / more realistic faker data — better mock values based on the schema context, accounting for all LLM providers, caching and customisation (requested by @IanVS, discussed with @stijnvanhulle)

[stijnvanhulle]