feat(Platform): allow DI override of userAgent

[waterplea]

Feature Description

Right now it is hardcoded as global navigator.userAgent:
https://github.com/angular/components/blob/main/src/cdk/platform/platform.ts

However in SSR there's a way to get userAgent from the request and accessing global entities is bad practice. Would be great if there was a token for userAgent like we do here, or maybe at least inject(DOCUMENT).defaultView?.navigator?.userAgent way of accessing it, rather than just global navigator.userAgent

Use Case

No response

[lazerg]

Worth noting that every navigator.userAgent read in there is short-circuited by this.isBrowser, which comes from PLATFORM_ID. On the server none of them execute and all the flags are already false, so a userAgent token by itself wouldn't change anything under SSR. To get value out of it you'd also have to drop the isBrowser gate, and that's the risky half: the flags would start coming back true during server rendering, and everything that consumes them (tooltip, input, slider, paginated-tab-header, interactivity-checker, text-field/autosize, overlay-outside-click-dispatcher, layout/media-matcher) uses them for event wiring, focus handling and DOM measurement that only happens in the browser anyway. New hydration-mismatch surface, no change to the rendered markup.

For the override itself, Platform is a plain root-provided service, so replacing it wholesale already works today and is how the CDK's own tests do it:

{provide: Platform, useValue: {isBrowser: true, IOS: true, WEBKIT: true, /* ... */} as Platform}

See focus-monitor.spec.ts. That gets you server-detected values without a new token.

The inject(DOCUMENT).defaultView?.navigator half seems like the more defensible ask on its own, since ViewportRuler, FocusMonitor and HighContrastModeDetector already reach the window through _document.defaultView || window rather than the global.

[waterplea]

isBrowser guard can be safely removed if the userAgent part was properly handled, for example it can be:

userAgent = inject(DOCUMENT).defaultView?.navigator?.userAgent || '';

And then all the regex tests would return false. Accessing window and CSS can also be rewritten as inject(DOCUMENT).defaultView etc.

[lazerg]

platform-server's DOCUMENT isn't a bare document. parseDocument builds it with domino.createWindow(html, url) and returns window.document, so defaultView is a live domino Window. It does carry a navigator (domino's NavigatorID), but its userAgent is the empty string, and there's no chrome or CSS on that window, so under the default server setup the regexes would come out false the way you describe.

Where it stops holding is the case the comment above isBrowser is about, a document shimmed without you knowing. platform-server can run with DOM emulation off against a global jsdom or happy-dom, and both of those ship a default UA along the lines of Mozilla/5.0 (darwin) AppleWebKit/537.36 (KHTML, like Gecko) jsdom/30.0.1. defaultView?.navigator?.userAgent picks that up, WEBKIT goes true on the server, and everything downstream of it branches the wrong way during rendering. Same class of false positive PLATFORM_ID was brought in for in #10659.

One more wrinkle: IOS uses !('MSStream' in window), and in throws on undefined, so that one needs an actual guard, optional chaining won't cover it.

Moving the global reads to inject(DOCUMENT).defaultView still looks worth doing to me, it lines up with what ViewportRuler and FocusMonitor already do. It just makes the reads safer rather than making the platform check redundant.