task_local context over context objects

- interner impls logically separate from API in orchid-base (default host interner still in base for testing)
- error reporting, logging, and a variety of other features passed down via context in extension, not yet in host to maintain library-ish profile, should consider options
- no global spawn mechanic, the host has a spawn function but extensions only get a stash for enqueuing async work in sync callbacks which is then explicitly, manually, and with strict order popped and awaited
- still deadlocks nondeterministically for some ungodly reason

orchid-base/src/msg.rs

@@ -6,7 +6,8 @@ use orchid_api_traits::{Decode, Encode};
 
 pub async fn send_msg(mut write: Pin<&mut impl AsyncWrite>, msg: &[u8]) -> io::Result<()> {
 	let mut len_buf = vec![];
-	u32::try_from(msg.len()).unwrap().encode(Pin::new(&mut len_buf)).await;
+	let len_prefix = u32::try_from(msg.len()).expect("Message over 4GB not permitted on channel");
+	len_prefix.encode_vec(&mut len_buf);
 	write.write_all(&len_buf).await?;
 	write.write_all(msg).await?;
 	write.flush().await
@@ -15,7 +16,7 @@ pub async fn send_msg(mut write: Pin<&mut impl AsyncWrite>, msg: &[u8]) -> io::R
 pub async fn recv_msg(mut read: Pin<&mut impl AsyncRead>) -> io::Result<Vec<u8>> {
 	let mut len_buf = [0u8; (u32::BITS / 8) as usize];
 	read.read_exact(&mut len_buf).await?;
-	let len = u32::decode(Pin::new(&mut &len_buf[..])).await;
+	let len = u32::decode(Pin::new(&mut &len_buf[..])).await?;
 	let mut msg = vec![0u8; len as usize];
 	read.read_exact(&mut msg).await?;
 	Ok(msg)