Use Rc::try_unwrap for (possibly?) better performance

[matheus23]

Putting this up here for discussion.

We can make use of mutability in Node::set_value and Node::remove_value, by turning values of type Rc<Node<...>> into mut Node<...> using Rc::try_unwrap(...).unwrap_or_else(|rc| (*rc).clone()).
Essentially that'll just give us an owned Node if we hold the only reference to it, or it'll clone the inner Rc.
Theory is that if the calling code is basically doing lots of changes on the HAMT in a tight loop, that's going to get sped up, since that'll end up being lots of in-place mutation.

The benchmarks don't look too promising though. I got ~10% improvements.
To test this, just compare the two commits in the PR against each other. Specifically their cargo bench -p wnfs-bench node.

Also, I had to hand-expand the async_recursion macro for set_value and remove_value, because otherwise there would've been lifetime issues with the original versions. We should probably figure out another way, but only if we really want to keep this PR.

[codecov]

Codecov Report

Merging #85 (60a4233) into main (80dbe82) will decrease coverage by 0.29%.
The diff coverage is 57.35%.

@@            Coverage Diff             @@
##             main      #85      +/-   ##
==========================================
- Coverage   66.05%   65.75%   -0.30%     
==========================================
  Files          26       26              
  Lines        1903     1907       +4     
  Branches      486      489       +3     
==========================================
- Hits         1257     1254       -3     
- Misses        248      253       +5     
- Partials      398      400       +2     

Impacted Files	Coverage Δ
wnfs/src/common/metadata.rs	67.56% <0.00%> (ø)
wnfs/src/private/hamt/node.rs	62.87% <58.73%> (-2.77%)	⬇️
wnfs/src/private/forest.rs	56.33% <66.66%> (ø)
wnfs/src/private/hamt/pointer.rs	49.12% <0.00%> (-1.76%)	⬇️

[matheus23]

@appcypher and I figured out that adding + 'static to K, V and H type parameters would make the lifetime annotations redundant.

It's for sure safe to annotate H: 'static, but for K and V it's not too obvious. It'd limit users of Node<K, V, H> to only put owned things (never references e.g. no &str for keys) into the HAMT. That's the case for all of rs-wnfs, so technically we could just do that? That'd make the code cleaner again and we could use #[async_recursion(?Send)] again.

[appcypher]

LGTM! 🎉

And like you suggested, maybe we add 'static for H at the impl level. Doesn't matter much either way.