[andreaTP]

Implements wasi:cli@0.2.12, the last interfaces a typical Rust wasm32-wasip2 program needs.

• Bindgen: a result used as a value (a parameter, or inside another type) now maps to a new run.endive.cm.runtime.Result<T, E>, lifted and lowered like tuples. exit: func(status: result) needs it. Results returned by a function keep their current mapping.
• WIT: deps/wasi_cli@0.2.12.wit is printed like the other deps, including its imports and command worlds, and the endive:wasi world is now include wasi:cli/imports@0.2.12.
• Hosts (run.endive.cm.wasi.p2.cli):
  • environment and arguments;
  • exit and exit-with-code, which throw WasiExitException;
  • stdout and stderr written synchronously to the embedder's streams, which are never closed, with a failed PrintStream such as a broken System.out pipe reported to the guest;
  • stdin through one reader per Java stream, shared by every host given it and started on the guest's first read or poll, with backpressure;
  • terminal interfaces controlled by per-stream TTY flags.
• API: the builder method names follow Endive's WASI p1: withArguments, withEnvironment, withStdin/withStdout/withStderr (each optionally with a TTY flag) and inheritSystemStreams. int run(ComponentStore, WasmComponent) calls the guest's wasi:cli/run export, matched to 0.2 by the same compatible-version rule as imports (Link guests importing semver-compatible interface versions #69), and returns its exit code.
• TTY default: streams default to not being a terminal, unlike p1, so captured output stays free of terminal styling. inheritSystemStreams counts them as terminals when the JVM has a console, which JDK 22 to 24 report even for redirected streams, as its Javadoc says.

Tests: CliGuestTest (12 end to end, WAT guest), unit tests for each host, the bindgen ResultValuesTest, a bindgen test of an error type one interface returns and another takes as a value (with #73's shared exceptions), and a virtual-thread run test in wasi-p2-loom-tests.

[jeremyg484]

Looking great!

I've reviewed it mainly with an eye towards compatibility with my in-progress wasi:sockets, and noted (mostly for myself since this will likely merge first) one case in the error handling in bindgen that will need to be handled carefully when we merge sockets.

A few other potential issues were caught with an AI review that feel like they're worth addressing in some form.