fix(cli): timeline lists clips nested in a timed element and lint flags ones that never show - #5732

Requesting changes at ef13e87b. The timing and the lint rule are right, but the new nested rows now also feed the timeline edit commands. As a result, the very edit the new warning asks for is refused.

Blocking

1. Edits on a nested clip, or on its wrapper, collide with each other (timeline/a2Shared.ts overlap and the duplicate branch of mutationConflict). Both checks compare against every row from allRows in the same file on the same trackIndex, and wrapper children are now in that list. A child without data-track-index falls to track 0, the same lane as a wrapper on track 0. With this PR's own fixture (#shot2 at 5–10s on track 0, holding #lost at 0–2s and #seen at 6–8s):

   • timeline move #lost 6, which is the fix the new warning suggests, is refused as "would overlap #shot2": the child collides with its own parent.
   • timeline move #shot2 4 is refused as "would overlap #seen". This worked on main, where children weren't rows. The suggested fix, "move the named neighbour", then points at the wrapper's own child.
   • timeline duplicate #seen inserts at 8, inside #shot2, so it's refused. The duplicate branch has no --overwrite bypass.

   Skipping ancestor and descendant rows (through parentRow) in both checks would fix it, along with a move test on a nested clip.

2. A sub-composition host inside a timed wrapper produces children: [null] (describeProject.ts flatten). add() recurses only into clip.inside, and a top-level clip's children are added only in the outer loop. A nested host's sub-composition clips are therefore never added, so at() returns undefined for them, and --json lists null children pointing at rows that don't exist. Example: <div id="shot" data-start="0" data-duration="5"><div data-composition-src="scene.html" data-start="0" data-duration="5"></div></div>. Fix: have add() also walk clip.children with host = clip, or don't read sub-compositions for nested hosts.

Non-blocking

• Nested <audio> gets the error, but it still plays. The runtime hides a nested element when its parent isn't showing (clipChainVisibleAt), but media playback uses only the clip's own window. A nested <audio> outside its parent's window is still heard. Lint says it "never appears", and Studio treats that as an error. Following the suggested fix would change when the audio plays. Excluding audio clips, or wording this for visual clips only, would avoid that.
• Lint ignores data-hf-media-start-basis="global". In a sub-composition, a legacy root-time media data-start is compared against a local parent window. The timeline gets this right through resolveMediaStartSeconds.
• In-point cuts stop at sub-compositions. cutRow cuts the sub-composition row but passes inside through unchanged. Wrapper children inside a host with an in-point keep their uncut times.
• The text view hides nesting. Wrapper children have nested: false, so the header counts them as top-level. The "no stable name" warning now prints on every unnamed nested row. kind/index refs also shift, which is worth a line in the release notes.

Checked and fine

• The body's timing claim holds. In startResolver.ts computeStart, a nested numeric data-start counts from the composition, not from the wrapper. Visibility also needs every timed ancestor visible (clipChainVisibleAt), so "never visible when the windows don't overlap" is the right rule for what shows on screen.
• resolveRowTiming describes inner rows with the same scope and origin. The overlap test matches the runtime's half-open clip windows.
• Both checks skip windows they can't resolve. A nested media clip with no data-start defaults to 0, which matches the compiler.
• isSubCompositionHost keeps both checks from walking into hosts. The new lint rule is its own entry in structureRules. nested, host and hostRow keep their meaning, and parentRow is purely additive.

CI is the only run. Not armed or merged.

— Rames

Approving at 071d68a8. Last time I requested changes at ef13e87b. Both blockers are fixed. The two pushes since fcc274f3 are behaviour-preserving: followsParentVisibility and isTimed replace the two inline guards with the same conditions, in the same order, and one test helper is inlined; 071d68a8 only undoes a stray reformat.

1. Nested edits no longer collide with their own parent. overlap and the duplicate branch of mutationConflict now skip rows in the same parent chain (sameLineage, walking parentRow). The new e2e cases run on the PR's own fixture. move #lost 8 succeeds, which is the move the warning suggests. move #shot2 4 and duplicate #seen also succeed. Sub-composition rows are in another file, so the file check still keeps them apart.
2. A sub-composition host inside a wrapper lists its clips. flatten's add() now walks clip.children with host = clip at every depth, so a nested host's clips become rows. The new test sees shot → [scene] and scene → [t1, t2], with no null children.

My non-blocking notes are mostly covered too:

• <audio> is now excluded, in both the lint rule and the CLI warning.
• Lint skips data-hf-media-start-basis="global".
• cutRow now carries the cut into inside, and there's a test for a clip nested in a wrapper.
• The header now counts wrapper children as nested.
• The text view prints only the never-visible warning, which also matches main, where it printed no warnings at all.

kind/index refs still shift for projects with wrapped clips. That's worth a line in the release notes, but it doesn't block.

CI is the only run. Not armed or merged.

— Rames