perf(server-renderer): optimize unrollBuffer by avoiding promises

[negezor]

Currently, unrollBuffer is an async function, which results in approximately a 10x slowdown in a mixed buffer and around a 1.5x slowdown in a synchronous buffer. Any Promise on the hot path is a performance killer.

In this PR, unrollBuffer will return a Promise only if a child buffer does; otherwise, it will always operate synchronously. However, even if a child buffer returns a Promise, we can continue to operate synchronously if further nesting is not required for another child buffer.

Of course, there are other functions where something similar happens, such as renderComponentVNode, but this requires separate consideration.

Performance was tested on the following hardware:
CPU: AMD Ryzen 7950x3D
System: WSL 2 Arch Linux on Windows 11
Node.js: v22.4.0

Before

 ✓ packages/server-renderer/__tests__/unrollBuffer.bench.ts (2) 1820ms
   ✓ unrollBuffer (2) 1819ms
     name             hz     min     max    mean     p75     p99    p995    p999     rme  samples
   · sync   3,526,611.07  0.0002  0.3646  0.0003  0.0003  0.0005  0.0006  0.0009  ±0.59%  1763307   fastest
   · mixed    303,890.63  0.0028  0.4434  0.0033  0.0032  0.0049  0.0055  0.0233  ±0.77%   151946

After

 ✓ packages/server-renderer/__tests__/unrollBuffer.bench.ts (2) 2321ms
   ✓ unrollBuffer (2) 2320ms
     name             hz     min     max    mean     p75     p99    p995    p999     rme  samples
   · sync   5,118,319.85  0.0002  0.0730  0.0002  0.0002  0.0002  0.0003  0.0005  ±0.11%  2559160   fastest
   · mixed  3,301,545.89  0.0003  0.2527  0.0003  0.0003  0.0005  0.0006  0.0009  ±0.48%  1650773

[github-actions]

Size Report

Bundles

File	Size	Gzip	Brotli
runtime-dom.global.prod.js	89.5 kB	34.2 kB	30.9 kB
vue.global.prod.js	146 kB	53.6 kB	47.8 kB

Usages

Name	Size	Gzip	Brotli
createApp	51.4 kB	20.1 kB	18.3 kB
createSSRApp	54.7 kB	21.5 kB	19.5 kB
defineCustomElement	53.7 kB	20.9 kB	19 kB
overall	65.1 kB	25.2 kB	22.7 kB

[yyx990803]

Tested this in https://github.com/eknkc/ssr-benchmark with renderToString and it got 20% faster, great job!