Database connect error during app boot should not terminate node process

Describe the bug

Steps to repro:

1. Scaffold new app (lb4 app)
2. Scaffold new mysql data source (lb4 datasource). Enter invalid connection info, so that the connection fails during boot.
3. Build and run the app

Expected:

Server starts successfully, and /ping endpoint may be ran.

Actual:

Server fails to boot and the process terminates

Alternative with lazy connect:

If "lazy connect" is enabled on the data source config, and a manual datasource.connect() is made later on, the error will cause an unhandled exception to crash the node server. Be aware that trying to run datasource.execute() directly without a connected data source will return a promise that will never resolve until a connection is made.

Note that this occurs even if the call is wrapped in a try/catch.
Further, normal crud repository fetchers, which internally queue the connect, are unaffected.

example:

{
   ...other datasource props,
   lazyConnect: true
}

@get('/test')
async test() {
  try {
    //execute will only queue, if not connected, causing an infinite wait
    if (!this.dataSource.connected)
      await this.dataSource.connect();

    //execute custom sql
    await this.dataSource.execute("select 1");
    return true;
  } catch(ex) {
    return false;
  }
}

Use Case:

Health check endpoints that check if parts of the application are online, including the database.

It is vital that the api continue to run, even if a dependency is unavailable. If a datasource.execute() or datasource.connect() fails inside a controller method, then it should not tear down the server with it!

Cli Version: 3.2.0

Logs

No response

Additional information

No response

Reproduction

repro provided above

Start with the datasource.connect() and datasource.execute() paths described in the reproduction, including their use during application boot and from the controller example. Reproduce the invalid MySQL connection with lazyConnect enabled, then verify that boot and controller-level failures are handled without terminating the Node process or leaving the request unresolved.