Guide developers to fix incidents in a way that is easily used by the LLM

[rszwajko]

Consider a use case of an organization that want's to migrate a large number of applications that use the same frameworks/patterns. One of the paths could be:

1. migrate in the "right" way one example application
2. use git history of this migration to train LLM for the remaining apps

The idea is that IDE should guide developers in that early stage. Some ideas:

1. encourage one incident in one commit approach.
   1. The advantage: no unrelated changes in the diff.
2. inject an identifier that would explicitly link a Kai suggestion with the commit (similar to Gerrit's Change-Id ) and possibly some meta-data. The advantages:
   1. no heuristic for guessing which commit fixed the incident
   2. information how the final fixed is different from the suggested fix
   3. meta-data for more context (see below)
3. if fixing an incident is a multi-step process then track the steps and attach them as meta-data. This can be implemented either as automated harvesting i.e. history of edit actions/refactorings/commands OR intentional action by the developer i.e. similar to CodeTour plugin. The advantages:
   1. the smaller the steps the better the suggestion from the LLM
   2. after each step user has chance to interact with the code i.e. cleanup the compilation errors and give the LLM a better starting point for the next step
   3. transparency: user can check how the given step was done in the original incidents and continue manually (or use the prompt)