`prefer-at`: inconsistent behaviour on arrays and strings

### Description Prior to #3200, `prefer-at` was a simplicity rule, suggesting replacing the complicated `array[array.length - 1]` (and other even more cursed incantations) with the nice simple `array.at(-1)`. It didn't, by default, suggest replacing `array[1]` with `array.at(1)`, since that is (slightly) more verbose. Now, `prefer-at` suggests replacing `"cow"[2]` with `"cow".at(2)`. I would suggest moving this behaviour behind the `checkAllIndexAccess` option, as it is for arrays.* \* In a new project, I wouldn't particularly mind banning indexing in favour of `.at`, especially for TypeScript without `noUncheckedIndexAccess` enabled. But I maintain our ESLint config for 500 000 lines of largely legacy JS/TS, where we can't really do that. ### Examples ```js // ❌ const foo = array[array.length - 1]; // ❌ const foo = 'string'['string'.length - 1]; // ✅ const foo = array[1]; // ✅ const foo = array.at(1); // ✅ const foo = 'string'[1]; // ✅ const foo = 'string'.at(1); ``` ### Additional Info _No response_