A BlobInputStream will start reading from the current position of the LargeObject used to create it, but the current/marked position does not account for this.
Initialising current position correctly will fix this.

Closes pgjdbc#3149

Using seek is more efficient than reading and discarding the bytes.
The only behavioural change is it now allows skipping past the natural end of the stream, but the method contract already allows for this. It should not break anything.

Closes pgjdbc#3144

BlobInputStream.seek will now throw if attempting to skip beyond the max size for a large object. This is a breaking change. The previous implementation was imperfect as it did not account for the server being compiled with a different limit.

The state of the stream after the exception is undefined, but hopefully sane. Further exceptions may be thrown on future operations.