Rework stream to use a conventional read loop

changelog/3216.feature.rst

@@ -1,3 +1,3 @@
 HTTPResponse.stream(amt=None) now yields data from non-chunked responses as it
 is available.  Previously all intermediate data was buffered and only returned
-as a single bytes object at the end of the the response.
+as a single, large, bytes object at the end of the the response.

src/urllib3/response.py

@@ -827,6 +827,8 @@ def _fp_read(
                 del data  # to reduce peak memory usage by `max_chunk_amt`.
             return buffer.getvalue()
         elif read1:
+            # work around: https://github.com/python/cpython/issues/112064
+            amt = amt or self.length_remaining
             return self._fp.read1(amt) if amt is not None else self._fp.read1()
         else:
             # StringIO doesn't like amt=None
@@ -1024,9 +1026,9 @@ def stream(
                     data = self.read1(amt=amt, decode_content=decode_content)
                 else:
                     data = self.read(amt=amt, decode_content=decode_content)
-
-                if data:
-                    yield data
+                if not data:
+                    break
+                yield data
 
     # Overrides from io.IOBase
     def readable(self) -> bool: