fix(postgres): don't panic if M or C Notice fields are not UTF-8 (l…

…aunchbadge#3346)

* fix(postgres): don't panic if `M` or `C` Notice fields are not UTF-8

This has been observed with an old version of PostgreSQL (11.0.4)
running on Windows Server 2016 with windows-1252 encoding and French
locale.

This change replaces invalid UTF-8 fields with a default string, so the
other fields can still be read if they are valid.

* Revert "fix(postgres): don't panic if `M` or `C` Notice fields are not UTF-8"

This reverts commit 362ca98.

* Check that Notice M and C fields are valid UTF-8

Otherwise, we return the invalid UTF-8 error to avoid panicking later.

sqlx-postgres/src/message/response.rs

@@ -153,13 +153,19 @@ impl Decode<'_> for Notice {
                 }
 
                 b'M' => {
+                    _ = from_utf8(&buf[v.0 as usize..v.1 as usize])
+                        .map_err(|_| notice_protocol_err())?;
                     message = v;
                 }
 
                 b'C' => {
+                    _ = from_utf8(&buf[v.0 as usize..v.1 as usize])
+                        .map_err(|_| notice_protocol_err())?;
                     code = v;
                 }
 
+                // If more fields are added, make sure to check that they are valid UTF-8,
+                // otherwise the get_cached_str method will panic.
                 _ => {}
             }
         }