Pull progress: split the stream on CR too

docker refreshes Downloading lines with \r, not \n -- readline()
parked them in the buffer, so the byte snapshots never surfaced and
the overall-progress dict stayed empty (the ⏳ summary never printed).
Chunked reads now split on both terminators; verified with a synthetic
\r-stream that the parser yields the refresh lines and computes the
52.2/65.1MB-style summary.

Co-Authored-By: Claude <noreply@anthropic.com>

evalharness/sandbox/prefetch.py

@@ -114,7 +114,7 @@ def _pull_one(image: str, verbose: bool = True) -> None:
         # per-layer lines; forward the informative ones (throttled).
         proc = subprocess.Popen(['docker', 'pull', ref],
                                 stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
-                                text=True, bufsize=1)
+                                bufsize=0)
         last_line = ''
         last_show = 0.0
         layer_prog = {}   # layer id -> (done_bytes, total_bytes): the raw
@@ -130,8 +130,33 @@ def _pull_one(image: str, verbose: bool = True) -> None:
                     return int(float(s[:-len(suf)]) * m)
             return int(float(s or 0))
 
-        for line in proc.stdout:
-            last_line = line.strip()
+        # docker progress lines end with \r (carriage-return refresh), not
+        # \n -- readline() buffered them until a rare \n arrived, so the
+        # Downloading snapshots (and the ⏳ summary fed by them) never
+        # surfaced. Split on BOTH terminators, chunked reads.
+        import io as _io
+        stream = _io.TextIOWrapper(proc.stdout, errors='replace')
+        buf = ''
+        _lines_iter = iter(lambda: None, 1)  # placeholder, replaced below
+
+        def _iter_lines():
+            nonlocal buf
+            while True:
+                chunk = stream.read(512)
+                if not chunk:
+                    if buf:
+                        yield buf
+                    return
+                buf += chunk
+                parts = buf.split('\r') if '\r' in buf else buf.split('\n')
+                if len(parts) > 1:
+                    for p in parts[:-1]:
+                        yield p
+                    buf = parts[-1]
+
+        for line in _iter_lines():
+            line = line.strip('\n ')
+            last_line = line
             m = _re.match(r'^([0-9a-f]{12}): Downloading.*?([\d.]+[kKMG]?B)/([\d.]+[kKMG]?B)', line)
             if m:
                 layer_prog[m.group(1)] = (_bytes(m.group(2)), _bytes(m.group(3)))