Replace hash_gid with FNV-1a

[fuzzypixelz]

hash_gid is called on Zenoh subscriber callbacks and shows up when profiling as a hot function. The use of std::stringstream seems to be relatively inefficient. A significant portion of hash_gid's running time is spend allocating & deallocating the std::stringstream.

[fuzzypixelz]

We seem to have many uses of std::unordered_map where the hashing is done manually instead of relying on the Hash template parameter. I'll revisit that before marking this pull request as ready for review.

[clalancette]

I left a question inline.

[clalancette]

So we are already using xxhash3 in liveliness_utils to take a string and hash it into a 128-bit quantity. Do we need to hash again? In other words, could we just use the 128-bit GID directly and use that as the key?

[fuzzypixelz]

I believe std::hash requires a size_t.

[Yadunund]

We talked about this today and one optimization to make is to rely on FNA-1a to store the hash of the liveliness string as a std::size_t here and send this size_t over any attachments as opposed to the 16B Gid.

[fuzzypixelz]

@Yadunund After taking a closer look at this, things are more complicated than they seem. FNV only outperforms XXH3 for very small keys (see https://github.com/Cyan4973/xxHash/wiki/Performance-comparison#benchmarks-concentrating-on-small-data). So it's better to simply move to a 64-bit version of XXH3, as long as we agree that GID can be 8 bytes instead of 16. What do you think?

[Yadunund]

This is looking good to me. A couple minor nitpicks to address after which this is ready to merge.