Automatically allocate a new chunk group instead of throwing error due to buffer size limits

[onurctirtir]

As in #5918, we might throw an error when writing into columnar storage due to several buffer size limits enforced by safestring library or by Postgres.

While the buffer size limit for memcpy_s() is 256MB, memcpy() doesn't enforce such a limitation; so #6419 attempts improving this situation by switching to use plain memcpy() instead memcpy_s(). Note that even if we use memcpy(), memory allocators --such as enlargeStringInfo()-- still enforces a limit of 1GB when writing a chunk group into the disk.

Indeed, we could almost completely remove such memory limitations by deciding to allocate a new chunk group in ColumnarWriteRow() instead of throwing an error in the runtime (i.e., when trying to expand the latest chunk group based on GUC limits).

[JelteF]

The workaround for this issue is to lower the columnar.chunk_group_row_limit setting. Possibly only for the dataload operation that is having trouble.

[EraYaN]

The workaround for this issue is to lower the columnar.chunk_group_row_limit setting. Possibly only for the dataload operation that is having trouble.

This doesn't work since the minimum is 1000 rows, so if values are larger than 1 MB it breaks anyway.