read_bytes(): use previous implementation again for small reads

For small reads, the new code that tries to avoid unnecessary reads is
noticeably slower than the previous code that reads unconditionally. In
the worst case (1-byte reads), the new code is 13 times as slow as the
previous implementation. The potential memory/IO savings only become
worth it for larger reads, where the performance difference disappears.

Co-authored-by: Petr Pucil <[email protected]>

Add ValidationExprError

See kaitai-io/kaitai_struct#435

Reverting change that broken ~13 tests

Refreshed links to point to actual resources

Pumped version

Pumped version