[REVIEW] CSV Reader: Fix a segfault when byte range aligns with a page

[vuule]

Fixes rapidsai/dask-cudf#102

• Remove the extra byte that was copied from the last chunk in launch_storeRecordStart(). The intent was to copy an extra byte for all chunks except the last one, but the implementation did the opposite. This cause a segfault when the map size aligns with a page.
• Remove the code to explicitly handle \r\n terminator in storeRecordStart() and countRecords(), since regular check (\n) can cover this case too.
  *When mapping the file, num_bytes is now updated to account for page padding in all cases.
  *Add a test for the failing case.

[j-ieong]

You need to fix the typo in the PR title as well.

[vuule]

You need to fix the typo in the PR title as well.

Fixed.