[code-infra] Add react-testing-library eslint plugin

[oliviertassinari]

configure eslint-plugin-testing-library and fix issues

See mui/toolpad#3648 as an implementation example.

Could also be helpful in:

• 1. [code-infra] In react tests, clean up our act() #174
• 2. Use screen at the import level, not the mounting point level:

-import { render } from '@testing-library/react';
+import { render, screen } from '@testing-library/react';

   test('renders content correctly', async () => {
-    const { getByText } = render(<AppProvider>Hello world</AppProvider>);
+    render(<AppProvider>Hello world</AppProvider>);

-    expect(getByText('Hello world')).toBeTruthy();
+    expect(screen.getByText('Hello world')).toBeTruthy();
   });
 });

For example mui/material-ui#41061 (comment)

• 3. Use user at the import level, not the mounting point level: [core] Support @mui/material@6 peer dependency mui-x#14142 (comment). Actually maybe not: [code-infra] Optimize @testing-library/user-event material-ui#43804 (comment).

[Janpot]

Now that we're experimenting with vitest, it looks like using global screen (and global cleanup) prevents our tests from running in parallel. We should actually do the opposite of their recommendation if we want to strictly isolate our tests 😄

[oliviertassinari]

Are they mounting multiple components on the same jsdom instance in the same JavaScript context? I would expect that Vitest uses multiple node processes, and hence it doesn't matter. Or can we assign the right jsdom instance to the screen API with a mutation?

[Janpot]

Are they mounting multiple components on the same jsdom instance in the same JavaScript context? I would expect that Vitest uses multiple node processes, and hence it doesn't matter. Or can we assign the right jsdom instance to the screen API with a mutation?

Vitest uses multiple processes, that works, but seems to be about 3x slower than mocha (will have to investigate further). I tried disabling their isolation but it acts too weird, even when disabling the per-file parallelism. Probably a dead end for now, yes...