Alright, I found the source of my refcounting bug. I updated the PR to include the C version of the dump for PyBytes.

I ran Serhiy's microbenchmarks on the C version and I could not detect any overhead on small bytes objects while I get a ~20x speedup (and no-memory copy) on large bytes objects as expected.

I would like to update the `write_utf8` function but I would need to find a way to wrap `const char* data` as a PyBytes instance without making a memory copy to be able to pass it to my `_Pickle_write_large_bytes`. I browsed the C-API documentation but I could not understand how to do that.

Also I would appreciate any feedback on the code style or things that could be improved in my PR.