`Client::login` should support retries in case of 429 / Too Many Requests

I'm using matrix-rust-sdk both in CI and for non-trivial bots that may require many clients.

Right now, Client::login uses a RequestConfig::short_retry(), ignoring any RequestConfig we may have passed in ClientBuilder::request_config, which makes the method fail if the server responds with a 429. This means that I need to manually retry my logins. It's definitely possible, but it's both ugly code and easy to get wrong.

I'd like the ability to have a Client::login that I can just trust to retry.

One possibility would be to introduce a method

impl ClientBuilder {
   /// Specify a `RequestConfig` to use during login, e.g. for retrying.
   ///
   /// By default, `ClientBuilder` assumes that its users wish to be informed quickly if they client
   /// cannot connect to a homeserver and will use the specialized `RequestConfig::short_retry()`
   /// to fail early.
   ///
   /// Use this method to e.g. enable further retries in case of 429 "Too Many Requests".
   pub fn login_request_config(self, RequestConfig) -> Self;
}

Caveat: In terms of API design, it's a bit odd.

Start at the Client::login and ClientBuilder entry points, then trace how RequestConfig::short_retry() is selected and how ClientBuilder::request_config is currently used. Resolve the API-design question around a login-specific configuration, and consider the 429 retry behavior the issue describes done when login can reliably apply the intended retry configuration.