Exposing esbuild's charset option to shrink non-ASCII critical scripts

[constantly-dev]

First off, thanks for maintaining this open-source project — it's a really focused, well-thought-out primitive.

I've been looking into how outputSizeLimit's byte budget gets spent, and ran into an esbuild default that eats into it without any way to opt out. Tested against esbuild@0.28.1 and @woowabros/vite-plugin-critical-script@1.0.0.

charset defaults to ascii, so esbuild escapes every non-ASCII character to a \uXXXX sequence. For a critical script embedding non-Latin text, this measurably inflates the byte count outputSizeLimit checks against. Compiling this line:

window.__t = { title: '안녕하세요 반갑습니다 오늘도 좋은 하루', items: ['한글', '테스트', '가나다라마바사'] }

comes out to 229 bytes with the default ascii charset and 142 bytes with utf8 — a 61% difference on the exact same source. Since the compiled script ends up inside a UTF-8-encoded HTML document either way, there doesn't seem to be a correctness reason to force the ASCII-escaped form.

This is an esbuild option the plugin already knows how to pass through, the same way target and define are exposed today. A possible API could look like this:

criticalScriptPlugin({ charset: 'utf8' })

Typed straight from esbuild (esbuild.BuildOptions['charset']) and passed to esbuild.build() unchanged, so the default stays exactly as it is today (ascii) and nothing about existing behavior changes.

Let me know if this direction makes sense. If so, I'm happy to put together a PR with the option and a test, mirroring how target is implemented.