fix(runtime): lay out <panel> children as a column

[SunkenInTime]

What

<panel> projected directly onto Native SDK's panel kind, whose children overlay each other. The contract (sdk/CONTRACT.md lines 127 and 604) promises column layout. Two <text> children in a <panel> rendered at identical bounds, one on top of the other.

Now <panel> lowers exactly like a styled <column>: outer panel for paint and effects, inner column for child layout and gap.

Receipts

• Before: experiments/render-loop-cadence/panel-probe/ (branch master today). Snapshot shows both texts at bounds=(8,8 56x80).
• After: same widget, texts at (32,31 7x11) and (32,48 7x17), matching the <column> control beside them.
• New unit test: panel lowers to a painted box around a column so children stack vertically.
• zig build test: 89 passed, 1 skipped.

Why it went unnoticed

weaver check passes, the receipt says ok, and the semantic tree has every node. Only pixels show it. One of four agents in the render-loop cadence experiment shipped this exact defect at 7/10 self-confidence.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[greptile-apps]

Greptile Summary

Native <panel> elements now place their children in a vertical column while retaining the panel as the painted container. Runtime checks verified stacked children, single gap ownership, retained alignment and wrapping settings, and preserved outer paint effects.

Confidence Score: 5/5

The panel layout change is safe to merge based on passing null-backend runtime checks and targeted before-and-after behavior checks.

The checks demonstrated the former direct-child panel structure, verified the new column structure, rejected duplicate gap ownership under a deliberate mutation, and confirmed that paint effects remain on the outer panel. The current null-backend runtime suite also completed successfully.

Files Needing Attention: No changed files need further attention. runtime/src/main.zig contains the validated projection and regression coverage.

T-Rex Logs

What T-Rex did

• I compared the former direct-child panel projection with the current nested-column projection and verified the null-backend runtime tests pass (88 tests passed, 2 were skipped, and none failed) with the test graph compiling.
• I validated that the gap is assigned to the inner column and cleared on the outer panel, and that restoring the implementation made the focused regression test pass.
• I inspected the mappings from retained panel options to the nested column and confirmed current runtime test coverage, noting a historical worktree could not start due to a missing Native SDK dependency.
• I ran equivalent paint-effect reproductions before and after the layout change, confirmed the outer panel's visuals stayed the same while the inner child changed to a column, and executed the panel-lowering regression test on the current checkout.
• I removed temporary validation edits from the runtime files and captured the final diff and repository status showing no staged or unstaged changes.

_{Ran code and verified through T-Rex}

_{Reviews (2): Last reviewed commit: "fix(runtime): lay out <panel> children a..." | Re-trigger Greptile}

[SunkenInTime]

Re the Greptile P1: the flagged setCrossAlign(panel, .center) call is not in this PR. The commit touches runtime/src/main.zig only at the .panel lowering arm and adds one test, panel lowers to a painted box around a column so children stack vertically, which uses setNumberProp(panel, "gap", 6) and no alignment setter. git grep setCrossAlign b48e600 -- runtime/src/main.zig returns nothing. The uncompilable test named in the review (panel inner column preserves child alignment and wrapping projections) is T-Rex's own scratch test, not part of the branch.

Local receipt on this exact commit: zig build test --summary all → 9/9 steps, 89/90 tests passed, 1 skipped, on macOS.

The failed macOS headless job was dev watcher normalizes atomic saves and ignores Weaver output (expected 1, actual 2), a filesystem-timing test in cli/test/dev-watcher.test.mjs that this runtime-only change cannot affect; the same job passed on #77 minutes later. Rerunning it.