write instead of write_all causes very rare failures

[mfro]

In the very rare cases where the underlying stream can only accept a small number of bytes for the first write call, the VarIntWriter functions will behave incorrectly, discarding the bytes that aren't written in the first call. I believe these calls should be replaced with write_all

[dermesser]

Thank you, I've implemented a backwards-compatible fix for this and released new versions.