Fix garbled JSON-lines output sometimes

In `kart diff -o json-lines --add-feature-count-estimate={value}`,
two threads would write to the output. This correctly used a lock to
prevent races, but there is an intermediate bytearray which is written
just beforehand (introduced with #1025) and this also needed to use
a lock to prevent races.

This change moves that intermediate bytearray under the same lock as
writing the output buffer, thus avoiding the race condition

Author: craigds

kart/json_diff_writers.py

@@ -247,10 +247,10 @@ def __init__(self, *args, diff_estimate_accuracy=None, delta_filter=None, **kwar
         self._output_buffer = bytearray()
 
     def dump(self, obj):
-        # https://jcristharif.com/msgspec/perf-tips.html#line-delimited-json
-        msgspec_json_encoder.encode_into(obj, self._output_buffer)
-        self._output_buffer.extend(b"\n")
         with self._output_lock:
+            # https://jcristharif.com/msgspec/perf-tips.html#line-delimited-json
+            msgspec_json_encoder.encode_into(obj, self._output_buffer)
+            self._output_buffer.extend(b"\n")
             self.fp.buffer.write(self._output_buffer)
 
     def write_header(self):