[Discussion Needed] Allow constructing Hashes from underlying byte array

[dongcarl]

With #14 merged, there is no longer a direct way to construct Hashes from an underlying byte array. This reintroduces a way to do so. This also allows for a more efficient implementation of impl<D: Decoder> Decodable<D> for sha256d::Hash.

Needs much discussion because of concerns brought up in #10.

[TheBlueMatt]

I'd personally vote for removing *::Hash::from_slice and skipping this, but that implies that rust-bitcoin will need to keep its Sha256dHash (at least) type, but I think most of the places rust-bitcoin would need to use that would never care about, eg, converting it to a secp private key, and maybe we can rename the type to something more descriptive given it may literally just be used for Bitcoin object hashes (ie block hash, txid...).

[apoelstra]

I think with associated constants we can make this more generic now.

[apoelstra]

I don't know what I meant by "more generic" above. This is fine.

[apoelstra]

I also think that having a dedicated method should make this reasonably hard to use by accident/incorrectly, unlike From which is supposed to be for borderline-free conversions and kinda encourages YOLO usage.