Agent supported rebase idea. Keeping a fork fresh.

[janb84]

One of the harder parts of maintaining a fork is to keep it somewhat in sync with upstream changes.
Often this sync process is a tedious endeavor that needs extensive involvement of humans. Now that we live in bold new world with large language models I think we can, and need to, utilize them to simplify this process.

In order to maximize the effectiveness of LLM's we need to re-structure the workflow of creating changes for the fork. In a nutshell each deviation (patch) form the master (upstream) branch needs to have, next to the changes, an invariant test and a SPEC.md so that the LLM has a good idea what a correct end state of the patch is. With this in mind, the LLM can independently mutate the patch-set to mitigate conflicting code changes upstream and end up in a new synced state.

Also to ease the maintenance of the fork further some other additional rules are introduces, see FORK WORKFLOW for full write-up.

Currently i've only converted 2140 PR4 to this new format, and I will convert some more to test the feasibility of the new workflow.

Happy to hear any feedback.

[janb84]

As an example I followed the fork workflow document and created:

A forkpoint from master (bc33509) to fork/bc33509ae2
A Patch from 2140PR4 [patch/01-remove-gui] https://github.com/janb84/bitcoin/tree/patch/01-remove-gui
A PR to apply the patch on the forkpoint janb84#1