Specialize extension payload encoding in Unparsed

Extension payload, after the RFC4251-encoded name string, is an opaque
blob, not an RFC4251 array of bytes, so it MUST NOT have a u32 length
header.

This had been handled in the `<Extension as Encode>::encode`
implementation, but that led to a bug in which there was a mismatch
between the number of bytes written and the length calculated. This
commit consolidates all the special-case handling of the opaque blob
into `impl Encode for Unparsed`. Any messages (i.e. Extension) that
contain `Unparsed` fields can now just call `encoded_len()` and
`encode()` like for any other field type.

Signed-off-by: Ross Williams <[email protected]>

src/proto/message.rs

@@ -622,10 +622,7 @@ impl Encode for Extension {
 
     fn encode(&self, writer: &mut impl Writer) -> ssh_encoding::Result<()> {
         self.name.encode(writer)?;
-
-        // NOTE: extension messages do not contain a length,
-        // as the inner Vec<u8> will be encoded with it's own length.
-        writer.write(&self.details.0[..])?;
+        self.details.encode(writer)?;
         Ok(())
     }
 }
@@ -657,7 +654,12 @@ impl Encode for Unparsed {
     }
 
     fn encode(&self, writer: &mut impl Writer) -> ssh_encoding::Result<()> {
-        self.0.encode(writer)
+        // NOTE: Unparsed fields do not embed a length u32,
+        // as the inner Vec<u8> encoding is implementation-defined
+        // (usually an Extension)
+        writer.write(&self.0[..])?;
+
+        Ok(())
     }
 }