Attempt to fix truncated Jedi LSP responses

This patch attempts to fix the truncated Jedit LSP responses which
produce warning messages about content-lenght mismatch. It does so
by looping around and trying again upon an empty read. The theory
is that the content producer is just not ready with more data yet.

- add function parameter 'max_empty_parts_in_a_row' with a default
value of 50.  This will be decremented on each empty read.
- change the default value of the 'max_parts' parameter to 100
since we should get larger chunks on each successful read.

Either of these values could be adjusted based on experience.

Author: dpriedel

python_packages/jupyter_lsp/jupyter_lsp/stdio.py

@@ -97,7 +97,7 @@ async def read(self) -> None:
                 self.log.exception("%s couldn't enqueue message: %s", self, message)
                 await self.sleep()
 
-    def _read_content(self, length: int, max_parts=1000) -> Optional[bytes]:
+    def _read_content(self, length: int, max_parts=100, max_empty_parts_in_a_row=50) -> Optional[bytes]:
         """Read the full length of the message unless exceeding max_parts.
 
         See https://github.com/krassowski/jupyterlab-lsp/issues/450
@@ -116,10 +116,11 @@ def _read_content(self, length: int, max_parts=1000) -> Optional[bytes]:
         raw = None
         raw_parts: List[bytes] = []
         received_size = 0
-        while received_size < length and len(raw_parts) < max_parts:
+        while received_size < length and len(raw_parts) < max_parts and max_empty_parts_in_a_row > 0:
             part = self.stream.read(length)
             if part is None:
-                break  # pragma: no cover
+                --max_empty_parts_in_a_row 
+                continue  # pragma: no cover
             received_size += len(part)
             raw_parts.append(part)