Padmé padding scheme.
Padmé leaks at most O(log log M) bits of information (with M being the maximum length of all messages) with an overhead of at most 12%, decreasing with message size.
This scheme leaks the same order of information as padding to the next power of two, while avoiding an overhead of up to 100%.
See Section 4 of Reducing Metadata Leakage from Encrypted Files and Communication with PURBs.
This function is meant to be used with
Padder, see this