Consider replacing xxhash with cryptographic hash function to not leak internal information

[tillrohrmann]

We should consider to replace our current hash partitioning algorithm (xxhash) with a cryptographic one in order to not leak internal information and to be not subject to targeted attacks by key/id forging.

[igalshilman]

Using a good hash function like SHA256 truncated to 64 bit, is a common practice, and has the following benefits:

• good distribution
• it is very fast to compute few GB/s of hashes is possible.
• has some secure properties (reduced due to truncation)

Specifically about distribution:

• very critical to our range based partition design, in addition having as little keys as possible that share the same partition key is useful in other places:
  1. storage_query a partition key is used as a low watermark in a symmetric hash join.
  2. in awakables we use the partition key for prefix scans to find an awakable with a specific invocation_id.
  3. invocation control - cancel/kill/inspect requires searching trough invocation statues with the same prefix.

From a quick servery Murmur is commonly used (Flink, Cassandra) and looking at SMHasher murmur and xxhash (which is currently used) are comparable.