host panic in stream support

[rvolosatovs]

On dffca0b

cargo test -p wasmtime-wasi-http 'p3::outgoing::http_0_3_outbound_request_large_post'

panics with:

thread '<unnamed>' panicked at /Users/rvolosatovs/.cargo/git/checkouts/witx-bindgen-80318ec43c150aef/40ca9b1/crates/guest-rust/rt/src/async_support/stream_support.rs:167:9:
assertion failed: buf.remaining() == 0 || matches!(status, StreamResult::Closed)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
test p3::outgoing::http_0_3_outbound_request_large_post ... FAILED

[alexcrichton]

I think I know what could cause this, will debug soon

[dicej]

We should also make sure the host isn't accidentally doing a zero-item read or write, which is easier to do as of my latest API changes (e.g. passing Vec::new() to StreamReader::read, which will see a capacity of zero and assume the caller wanted to do a zero-byte read).

[alexcrichton]

"fixed" in bytecodealliance/wit-bindgen#1259, although the test is infinite looping (maybe in the zero-length writes @dicej mentioned), and I haven't debugged the infinite loop

[dicej]

As I suspected, the host is doing a zero-length read (presumably by accident). I've rebased feat/http onto main, updated the code to match, added nonzero-capacity assertions for each StreamReader::read call, and pushed the result here: 7b6abcb

I haven't dug into why the code is passing a zero-capacity buffer to StreamReader::read, so I'm not sure what the correct fix is. Note that StreamReader::read never grows the buffer it is given (even though e.g. BytesMut is capable of growing); it assumes the caller wants to read no more than the capacity. One possible fix would be to call BytesMut::reserve before StreamReader::read to reserve some capacity, but I'll defer to @rvolosatovs about the correct behavior here.