Implement .netrc support

[truelecter]

Please note, despite all my tries, I suck at rust and rust code was written with Claude 4.8 assistance. I cleaned excessive comments, simplified some functions it overengineered a bit and to my eye code looks reasonable.

I tested it myself and currently using this fork on my configuration.

Implements #23

[NotAShelf]

Thank you for the PR and the AI disclosure. I've left one quick comment. Other than this please:

1. Reword your commits to patch the repository conventions (scoped commits)
2. Add a quick section under documentation about the usage.

[truelecter]

Restructured commit messages, added .netrc note under Upstream Authentication README section. I did not include an example of the netrc file, but I'll add it if needed

[NotAShelf]

The guard (specifically !upstream.username.is_empty()) skips netrc lookup only when username is set. If only password is explicitly configured (no username), netrc will silently overwrite it anyway. We should also check upstream.password.is_some().

[truelecter]

Adding an upstream.password.is_some() guard here would result in effectively empty credentials, since password-only (or username-less) credentials are being filtered out at:

• ncro/crates/server/src/lib.rs

  Lines 387 to 392 in eb67df3

  	state
  	.upstreams
  	.iter()
  	.chain(state.fallback_cache.iter())
  	.find(|u| u.url == url && !u.username.is_empty())
  	.map(|u| (u.username.clone(), u.password.clone()))

• ncro/crates/router/src/lib.rs

  Lines 302 to 307 in eb67df3

  	let mut map = self.inner.upstream_auth.write().await;
  	if username.is_empty() {
  	map.remove(&url);
  	} else {
  	map.insert(url, (username, password));
  	}

The docs and the upstream struct also imply that password is optional while username is required (for auth, at least).

I didn't add a password check initially because password-only configs were silently ignored before. I think it'd be better to issue a warning here and fall back to the username/password from .netrc to at least try to authenticate.

Or did you mean that an existing password shouldn't be overwritten by .netrc, and should instead give precedence to the one set in the config?