docs: close out the three-way review backlog

[oratis]

F8, F9 and F15 landed in #229, #230 and #231, so §7's table has no open findings left.

Second-batch discoveries, recorded

• The event-envelope bug found in feat(desktop): Changes panel — review findings and the working tree #220 was the narrow case of a broader one: every item.completed without a payload.message was dropped on resume — approval, ask_user, error, review_finding. The app-server persisted them faithfully and the UI never showed them again.
• The protocol never carried reasoning at all, only reasoningTokens. The desktop wasn't a projection bug; there was nothing on the wire.
• It could start, read and resume a thread but not enumerate one, which is why the desktop grew a second reader of the session directory.
• Resolving a rebase conflict by keeping both sides needs a structural check. Two regex merges ate a CSS closing brace and a describe() terminator; "no conflict markers left" doesn't mean the file parses. CI and Playwright caught both.

Still open, written down rather than implied

• The sidebar's archive/delete buttons still go through Tauri. thread/archive is served but isn't the path the button takes — the other half of removing the second reader.
• The Linux bwrap sandbox has not had the empirical pass macOS got in feat(sandbox): make it a real axis, and turn it on by default #226; it only shares the mode resolution.
• Image input stays a scaffold on purpose.

Doc-only.

🤖 Generated with Claude Code