perf(angular): fix CLS regression from critical-CSS inlining + drop rxjs from the bundle

[nigrosimone]

Summary

Two independent optimisations to the Angular app:

1. Fix a CLS regression caused by Angular's critical-CSS inlining worth +23 Lighthouse performance points.
2. Remove rxjs from the shipped bundle by moving rxResource to resource().

1. CLS: disable inlineCritical

Angular's critical-CSS inlining (beasties) runs against the built index.html, where
<app-root> is still empty. It therefore only finds :root / html / body rules to
inline, and defers the real stylesheet:

<link rel="stylesheet" href="styles-*.css" media="print" onload="this.media='all'">

The app paints unstyled, then the stylesheet arrives and re-lays out the whole page.
Lighthouse attributes a 0.724 layout shift to <main class="main">.

Since nothing renders before the JS bootstrap anyway, making the stylesheet render-blocking
costs nothing on FCP/LCP and removes the shift entirely.

Metric	before	after
Performance	77	100
CLS	0.724	0
FCP	0.4 s	0.4 s
LCP	0.5 s	0.5 s
TBT	10 ms	0–10 ms
Speed Index	0.4 s	0.4 s

Accessibility and best-practices stay at 100.

2. Drop rxjs from the bundle

resource() is stable public API in Angular 22 (@publicApi 22.0) and takes a promise
loader, so @angular/core/rxjs-interop and the rxjs operators are no longer reachable:

• weather-state.service.ts: rxResource({ stream }) to resource({ loader })
• weather.service.ts: Observables to async/await over the existing fetch calls

rxjs stays in package.json it is still a required peer dependency of @angular/core but its operators no longer end up in the output.

	before	after
main JS (raw)	168.17 kB	158.64 kB
main JS (transfer)	48.65 kB	45.86 kB
Initial total (raw)	185.57 kB	176.04 kB
Initial total (transfer)	52.03 kB	49.24 kB

Bonus: the loader now forwards resource()'s abortSignal down to fetch, so switching
city actually cancels the in-flight request instead of letting it race.

eslint.config.mjs gains AbortSignal: 'readonly' alongside the other browser globals.