Improve fuzzability

[kerhong]

Fuzzing of the library could help find inconsistencies between parsing and encoding payloads, and also panics reachable from inputs.

The most generic fuzzer I can think would be:

• generate opcode, generate size, generate u8[size - opcode_size]
• try to parse
  • if read whole buffer and was successful in parsing
    • encode it
    • verify original buffer is same as encoding output

Fuzzing is usually a separate crate that depends on your crate, so it can only access pub items.

This needs:

• ability to parse from buffer and get total read size
  • could be done by exposing modified read_opcodes as pub fn read_opcodes(opcode: u16, body_size: u32, r: &mut &[u8]) -> Result<...>
• ability to write a header of any opcode + size
  • could be done by exposing write_unencrypted_server
  • can easily be written as part of fuzzing implementation, but would be more correct if it's the exact implementation as in the lib

from #77

[gtker]

As mentioned in #77 I would like to keep the library as simple to use as possible, so I don't want to expose unnecessary implementation details.

However, I might go along with exposing the required functions in a separate fuzzing module, or another such indirect manner.

It might also be valuable to implement Arbitrary for the opcodes and then generate full opcode structs. That way we can at least test the Write -> Read loop.

Couldn't it also work by just reading arbitrary bytes using ServerOpcode::read_unencrypted?