hash_combine results in hash collision

[suranap]

The current hash_combine caused a ShardingID collision when I ran with 14 nodes on Perlmutter. It looks like the first version that was in Boost. Here's some info on Boost's recent implementation.

These machine views produced the same hash of 1804451222.

MachineView<0=[44:46:1]>
MachineView<0=[41:97:1]>

I updated my branch to use the 64bit version. It might be better to copy their whole implementation, which dispatches to 32 or 64 bits depending on size_t.

[lockshaw]

Sounds like a great idea! Do you have time to file a PR updating hash-utils.h with the full implementation?

[lockshaw]

@goliaro Does this change also need to get merged into inference? What's the plan for now for merging into master vs inference--should all PRs until repo-refactor is merged go to master or inference (I'm guessing master, but just want to double check)?