serial: bidirectional TCP sink (mode = "tcp")

[codewiz]

Adds a fourth [serial] mode bridging Paula's serial port to a host TCP listener, like UAE's TCP: serial device (same default port, 1234; override with [serial] listen = "host:port").

With Mount AUX: + NewShell AUX: on the Amiga side, a connected client gets a full remote AmigaDOS console -- handy for scripted testing and for driving the emulated machine headlessly. One client at a time; a new connection replaces a finished one, and output with no client connected is dropped like an unplugged cable.

Implementation notes:

• A background thread owns the accept loop and the socket read half, staging input bytes through a channel with an atomic fill counter, so Paula's idle fast path (has_pending_input) stays a cheap atomic load and never touches a socket.
• Writes go directly to the client stream with TCP_NODELAY for shell interactivity.
• Zero new dependencies (std::net only), so it works on all platforms, unlike a pty-based bridge.
• Unit test covers the input/output round trip; tested interactively against an AUX: shell with minicom (via a pty bridge) and scripted clients.

🤖 Generated with Claude Code

[codewiz]

I have regrets about using TCP: I couldn't get a working Amiga shell with nc nor with screen, probably due to mismatches between VT100-ish terminal escapes and the ANSI escapes used by the shell.

Amiberry / WinUAE also relay the built-in serial to localhost:1234, which is probably a good choice for file transfers and modem emulation, but I think we should also have a pty mode that works directly with minicom -D /tmp/amiga-serial.

[Copilot]

Pull request overview

This PR adds a new [serial] mode = "tcp" option that bridges the emulated Paula serial port to a host TCP listener (defaulting to 127.0.0.1:1234, configurable via [serial] listen), enabling bidirectional I/O suitable for remote AmigaDOS AUX: shells and headless/scripting workflows.

Changes:

• Introduces TcpSerialSink implementing a bidirectional TCP serial bridge with a background accept/read thread and a fast has_pending_input() probe.
• Extends configuration parsing/serialization to support SerialMode::Tcp and the optional [serial] listen address.
• Updates the launcher’s serial mode labeling/cycling to include the new TCP mode (in MIDI-feature builds).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
src/video/launcher.rs	Adds TCP to the launcher’s serial mode cycle/label list.
src/serial.rs	Implements the new bidirectional TCP serial sink and a unit test for round-tripping.
src/emulator.rs	Wires SerialMode::Tcp into serial sink construction using the configured/default listen address.
src/config.rs	Adds SerialMode::Tcp, [serial] listen, and parsing/error-message updates.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[codewiz]

Thanks for the review - all four addressed in 582d7cf:

• buffered unsigned underflow (both comments): switched to AtomicIsize, so a read_byte that consumes a byte before the reader thread's matching fetch_add dips to a transient -1 debt instead of wrapping to usize::MAX and getting stuck-true.
• nc hint mangles IPv6: now formats local_addr.ip()/local_addr.port() separately (and uses the resolved local_addr, so port-0 binds print the real port).
• launcher drops [serial] listen on re-save: MachineSetup now carries serial_listen through from_raw/to_raw; added a round-trip regression test.

[LinuxJedi]

Sorry @codewiz, just merged a bunch of things and it caused a merge conflict. Can you sort that and then I should be able to get this in.

[codewiz]

No problem - rebased onto current main (the conflict was just the new [audio] lines landing next to the [serial] block in MachineSetup::to_raw). Builds clean, fmt/clippy happy, serial tests pass. Ready when you are.

[LinuxJedi]

Excellent stuff!