minor - simplify multi-byte unicode char in log message (pytorch#1548)

Summary:
Pull Request resolved: pytorch#1548

As title. The log message was (inadvertently?) encoding a ' as a
multi-byte ’ sequence (fancy unicode codepoint 8217 /
U+2019 RIGHT SINGLE QUOTATION MARK -
https://www.codetable.net/decimal/8217). This can make searching for
the log message harder to match. Switch to a simple ascii apostophe
(codepoint 39).

Reviewed By: bigning

Differential Revision: D51726626

fbshipit-source-id: 5e1ecb9ea4eab9ab4f9a0304d53c635015176531

torchrec/distributed/embedding_lookup.py

@@ -390,7 +390,7 @@ def prefetch(
                 ):
                     logging.error(
                         "Invalid setting on SplitTableBatchedEmbeddingBagsCodegen modules. prefetch_pipeline must be set to True.\n"
-                        "If you don’t turn on prefetch_pipeline, cache locations might be wrong in backward and can cause wrong results.\n"
+                        "If you don't turn on prefetch_pipeline, cache locations might be wrong in backward and can cause wrong results.\n"
                     )
                 if hasattr(emb_op.emb_module, "prefetch"):
                     emb_op.emb_module.prefetch(