PackageFileToCsv, PackageArchiveToCsv sometimes loses data when using big mode CSV generation

[joelverhagen]

The dropped data always relates to a single bucket.
Sometimes the dropped data is existing records at the beginning of the blob.
Sometimes the dropped data is existing records at the end of the blob.

Using the bucket range processing mode on the bucket with dropped records fixes the issue.

When this problem occurs, sometime aggregated telemetry is missing.

[joelverhagen]

This appears to be caused by truncated download of the existing CSV blob. In one example, the blob is 7 MB (as seen by the worker) but only 5 MB gets downloaded. I will add more telemetry for this.