Enforce read before write

[atirut-w]

This is something that Claude Code enforces, and it makes sure that in the event that 1) the AI makes changes to a file, 2) the user then edit the file further and 3) request the AI to make more changes, the AI would be forced to read the file again to "sync up".

Because this isn't enforced with Roo, and editing through diff isn't reliable on some models (i.e. GPT 4.1), what ends up happening is that the AI, instead of building upon the user's further edits, builds upon what it wrote prior to the user making more changes to the file.