Before this fix, an unnecessary memset was called in
sql_digest_storage.memset().

This affected:
- the server under normal operations,
  when collecting digests for a query,
  with typically an extra memset(1024) for each query.

- the performance_schema.events_statements_current,
  _history and _history_long tables, during a SELECT,
  with typically an extra memset(1024*1024) for each row.

This memset is un necessary, because the code making
copies of struct sql_digest_storage, or computing md5 of it,
is already only using the first m_byte_counts initialized bytes
and not the full array.

The fix is to simply remove the memset().