Fix offset calculation bugs in ObjectFetcher

[teor2345]

There are multiple bugs in the ObjectFetcher’s piece and segment offset calculations:

• we’re using segment piece positions (including parity pieces) in calculations with source piece counts and lengths
• we subtract 2 for optional padding, but if the mapping is right at the end of the segment, and the padding isn’t there (or the mapping is malicious), that calculation can underflow

I think this previously worked in automated and manual tests because the mappings were all in the first source piece, and never in the last 2 bytes of the segment. I’ll follow up with a PR containing tests for those situations.

Code contributor checklist:

• I have read, understood and followed contributing guide

[nazar-pc]

I would certainly prefer to rewrite it in terms of regular positions somehow, but this will work for now

[teor2345]

I would certainly prefer to rewrite it in terms of regular positions somehow, but this will work for now

I’ve added this to the list of cleanups in #3106