[joshiste]

Summary

SteadyBuddy can now save and run experiments (platform PRs #1995, #1997, #2129). The SteadyBuddy page still said it is read-only, which is wrong for every tenant that has the capability.

• How SteadyBuddy Works: the read-only warning is replaced by the actual safeguard. Every save or run is asked first. Auto mode skips only the question, never the permission checks.
• Create Experiments from Scratch via Chat: documents the save card (Yes, save it / Yes, save and run it / Decline) and what a decline leaves behind. Also covers the run offer after a plain save, running an existing experiment by key, the run card with its stop button, and the message at the end of a run. The draft-only flow stays documented for tenants without the capability.
• New subsection "Approve or Let SteadyBuddy Act on Its Own": covers the Manual / Auto mode switch, Yes, and switch to auto mode, a warning about real faults, the fact that auto mode still won't accept a run offer on its own, and how the mode is remembered.

Labels are quoted from the platform code (@Approval titles in PrimaryCapabilityTools, ApprovalOptionVO, ChatModeSwitch, AUTO_MODE_EXPLANATION). docs_lint.py reports no new findings.

Before merging

• Merge with the platform release that ships steadybit/platform#2172, which removes the AI_WRITE_TOOLS feature flag. This PR no longer describes the capability as tenant-dependent, and it no longer has the draft-only fallback paragraph.
• No new screenshots yet. chat-create-experiment.png still shows a draft card with its own Open/Run buttons. Since #2172, those buttons only appear after a declined save. The approval card and the mode switch would each deserve a screenshot too.

BusinessMap: https://steadybitgmbh.kanbanize.com/ctrl_board/2/cards/16474