refactor: Use unsigned ignore() consistently #24194

Currently CDataStream::ignore takes a signed value. However all other ignore() functions take an unsigned. This causes integer sanitizer issues when an unsigned value is passed to CDataStream::ignore.

Fix this by accepting unsigned in all ignore() functions consistently.

Also, simplify the code to remove the confusing "nSize negative` exception and fallback to the more meaningful "end of data" exception.

Sorry, this is wrong.

What was wrong about it?

I accidentally (partially) reverted #7933 (I think)

I don't see how. nSize is unsigned there (and non-negative).

It is unsigned in compressor, but will be made signed in streams. A sufficiently large unsigned value will become negative once signed. The throw-on-negative is one way to avoid integer overflow later on. The throw-on-too-large alone is not sufficient to catch overflow.

While overflow could still happen theoretically in the current code, I don't think it can practically. Unless there is really bad disk corruption.

I think there are two solutions:

• Keep the code as-is, and document why the unsigned->signed conversion is intended
• Make compressor throw directly instead of calling s.ignore(nSize) (and hoping for a throw), when nSize is larger than the maximum block size or so.

Created #24231 as third alternative