Release the flux-network changes as 0.2.0

[Bronek]

DRAFT: FURTHER REFACTORING COMING, BELOW WILL BE SOON OBSOLETE

This PR builds on #145 and changes only the workspace version in Cargo.toml and the derived workspace-package versions in Cargo.lock.

The networking changes since v0.1.3 break public API compatibility, so Cargo's pre-1.0 compatibility rules require a minor version bump. Flux uses one version for every workspace crate; the complete workspace therefore moves from 0.1.3 to 0.2.0.

Closes #143.

Release highlights

• StreamNetwork supports TCP and Unix-domain endpoints through one connection lifecycle, with network-owned or caller-owned polling.
• Multiple independently configured HttpService instances can share one network and poll.
• HTTP gains pull-based events, immediate or deferred responses, request deadlines with failure reasons, connection caps, lingering close for rejected uploads, and allocation-free warm paths.
• listen reports the endpoint it bound, including a TCP port selected by the kernel.

HTTP server example

use flux_network::{
    http::{HttpConfig, HttpEvent, HttpService},
    stream::{ConnectionGroupConfig, Endpoint, Framing, StreamNetwork},
};

let mut net = StreamNetwork::default();
let group = net.add_group(ConnectionGroupConfig {
    framing: Framing::Raw,
    max_connections: Some(256),
    ..Default::default()
});
let mut http = HttpService::new(&mut net, group, HttpConfig::default());
let bound = http.listen(&mut net, Endpoint::Tcp("127.0.0.1:0".parse()?))?;
println!("listening on {bound}");

loop {
    net.drive(None, &mut [http.as_service()], |_| {});
    while let Some(event) = http.next_event(&mut net) {
        if let HttpEvent::Request { responder, .. } = event {
            responder.respond(200, &[("content-type", "text/plain")], b"ok");
        }
    }
}

StreamNetwork::default() owns its poll and drive(None, ...) blocks until I/O or the next deadline. Applications that already own a mio::Poll can instead construct the network with with_registry and call next_deadline, handle_event, and tick from their loop.

Migration from 0.1.3

The released v0.1.3 networking API was TCP-only. Its users need the following source changes:

0.1.3	0.2.0
flux_network::tcp	flux_network::stream
TcpNetwork, TcpEvent, TcpGroup, TcpGroupConfig	StreamNetwork, StreamEvent, ConnectionGroup, ConnectionGroupConfig
listen(group, SocketAddr), connect(group, SocketAddr)	listen(group, Endpoint), connect(group, Endpoint) using Endpoint::Tcp(addr) or Endpoint::Unix(path)
listen(...) -> io::Result<()>	listen(...) -> io::Result<Endpoint>; TCP port 0 returns the selected port
Event field peer_addr: SocketAddr	peer: Peer, matched as Peer::Tcp(addr) or Peer::Unix
TCP options directly on TcpGroupConfig	ConnectionGroupConfig::tcp: TcpOptions with the same defaults
Default group name "tcp"	"stream"; callers relying on default log labels or telemetry suffixes will observe the change

TcpConnector, TcpStream, and TcpTelemetry keep their names but move from the tcp module to stream.

There is no HTTP migration from v0.1.3: that release did not contain the HTTP module. HttpService and its resilience features are new in 0.2.0.

Verification

The release checks pass with the updated lockfile:

cargo check --workspace --all-features --locked
cargo test --workspace --all-features --locked

Base: #145.

Assisted-by: Claude Code:claude-fable-5
Assisted-by: Codex:gpt-5