Make the rendering test context tracked

[NullVoxPopuli]

RenderingTestCase now declares its context properties tracked, so the suite stops depending on a legacy read path that RFC 957's end state removes.

Groundwork for emberjs/rfcs#1203 ("Deprecate this.get and this.set on test contexts"): the framework's own suite is the largest consumer of the pattern that RFC deprecates.

Why

Test state reaches templates through the top-level component, and tests mutate it with set(this.context, ...):

this.render(`{{#let (array this.personOne) as |people|}}...{{/let}}`, { personOne: 'Tom' });
runTask(() => set(this.context, 'personOne', 'Yehuda'));
this.assertText('Yehuda');

That drives a rerender only because a non-tracked property read still entangles a per-property tag. _getProp pays for that on every property read, and removing it is one of the levers measured in #21520. Today the framework's own suite is a large consumer of that path, with 771 set(this.context, ...) call sites across 43 files.

I have not yet been able to measure how many #21520 failures this accounts for. That branch aborts partway through its own suite, so the two runs I have cover different amounts of it and are not comparable. The claim here is only the one verified below: this is behaviour-neutral on main.

What

this.context is Component.extend(attrs), so the fix is one place rather than the 771 set(this.context, ...) call sites. Context properties are declared with tracked({ value }), which keeps every existing set call reactive under modern semantics. Values that are already decorators (computed properties, injections) bring their own reactivity and pass through untouched.

No behaviour change here: 9448 tests, 0 failures.

Not in this PR

runTask is the other half of the same migration and is harder. 1678 call sites across 84 files assume rendering flushed synchronously:

runTask(() => set(this.context, 'x', 'y'));
this.assertText('y');   // no await

Under the async scheduler it has not flushed. Rewriting those to await turns most of the suite async. The cheaper route is redefining runTask on a flushSync-style escape hatch, which does not exist yet. Worth settling that before touching call sites.

🤖 Generated with Claude Code

[NullVoxPopuli]

this is garbage, claude lol not what I asked