Fix add WebGPU renderer support in PerfHeadless

[YusefNsar]

Summary

Adds compatibility with Three.js WebGPURenderer in PerfHeadless, while keeping full backward compatibility with WebGLRenderer.

Detection

WebGPURenderer.getContext() exists but returns null (unlike WebGLRenderer which returns a WebGL2RenderingContext). Detection is done by checking the return value and the presence of getExtension/getParameter on it:

const glContext = gl.getContext?.() as WebGL2RenderingContext | null
const isWebGPU = !glContext || typeof glContext.getExtension !== 'function' || typeof glContext.getParameter !== 'function'

Changes

• GPU timing: EXT_disjoint_timer_query_webgl2 is WebGL-only. When WebGPU is detected, gl is not passed to GLPerf (skipping the WebGL timer query path). Instead, info.render.timestamp (written by WebGPU's built-in timestamp queries) is used as the GPU time value via logger.gpu || glRender.timestamp.

• Timestamp resolution: After each frame on WebGPU, resolveTimestampsAsync('render') is called so that info.render.timestamp is populated before the next paramLogger invocation.

• Renderer info: WEBGL_debug_renderer_info and ctx.getParameter are WebGL-only. On WebGPU, the infos block falls back to static 'WebGPU' strings. GPUAdapter.requestAdapterInfo() would provide real values but is async — left as a TODO.

• Draw call stats (calls, triangles, points, lines): gl.info.render is available on both renderers with the same field names, so these work unchanged.

What still doesn't work on WebGPU

• deepAnalyze / program list panel: relies on gl.info.programs which is WebGL-only and not present on WebGPURenderer.
• GPU vendor/renderer strings: requires the async GPUAdapter.requestAdapterInfo() API.

Test plan

• Verified FPS, CPU, GPU metrics display correctly with WebGPURenderer + enableTimestamps: true
• Verified no regressions on WebGLRenderer
• deepAnalyze gracefully shows nothing (empty program list) on WebGPU without throwing

[YusefNsar]

This is a fix for issue #63