Update MultiPL-E to v3 prompts

I've put together a new MultiPL-E release, v3 and this updates the harness to point to the new prompts.

There are fixes to old prompts as well as support for new languages, both described on the MultiPL-E dataset page. The nature of fixes is that some PLs have their scores go up a little. I've tested with with StarCoder2-15B (using VLLM) and will continue testing other models.

Author: arjunguha

bigcode_eval/tasks/multiple.py

@@ -80,7 +80,7 @@ class GeneralMultiPLE(Task):
 
     DATASET_PATH = "nuprl/MultiPL-E"
     DATASET_NAME = None
-    DATASET_REVISION = "d23b094346c5dbda1080a74bb2a24c18adbf7409"
+    DATASET_REVISION = "8a4cb75204eb3d5855a81778db6b95bfc80c9136"
 
     def __init__(self, language):
         self.language = language