Improving the use of stdlib and const-correctness where possible

[DiamonDinoia]

No description provided.

[magland]

@DiamonDinoia It's possible that these changes slowed down the decode. This is from the CI artifact for this PR:

Compared with the current timings:

Whereas for encode, things seem to have gotten better. Here's this PR:

Compared with current encode:

Could you try some benchmarks on your end, please?

[DiamonDinoia]

Sure, I will have a look

[DiamonDinoia]

I am not sure why there is a slowdown. As I don't see the same behavior locally. Maybe a difference in the stdlib or compiler version?

I propose we go through all the changes I have planned and if the slowdown persists we can revert the changes here.

[DiamonDinoia]

Also, I think the faster results might be a lucky draw. Even blosc is over 5GB/s in master but 4.8GB/s in this PR.