[spike] Fine-tuning options for LLMs

[codificat]

As part of #8, in order to adapt the language models we use to the domain we work on (ROSA), explore and document the various options available:

• prompt engineering with context
• model fine-tuning

For each option we want to understand:

• its pre-requisites (e.g. type, quality and quantity of data required, level of model access, cost of the adaptation process)
• pros and cons
• what OSS options are recommended for each (see [Research Spike] Collect open source implementations and frameworks of QA systems #9)