These unwraps and indexing (which hide more panic paths) irritated me, and also
would need to be tweaked once we get rid of the Hash::from_slice methods. I
figured I'd preemptively get rid of them.

I don't *think* it's possible to create a rangeproof with a sidechannel smaller
than 64 bytes (if you create a 0-sized "proof of exact value" then unwinding
will fail entirely, and anything larger I think has at least one ring, so 128
bytes or more). Unsure.

But better not to assume this by indexing recklessly into the sidechannel
message.

We should redo the error types here as well at some point.

Currently we allow decoding segwit v0 programs which have uncompressed/hybrid
keys (not allowed) and I suspect that if you provide a too-short address then
you'll get a panic here.

Lol, this code was so close and yet so far.

pset

6fee387 Bump version to 0.26.2 (Philip Robinson)
4428361 pset: fix slicing in KeySource::deserialize (Philip Robinson)
6e082a9 address: do a better job slicing bech32 data (Philip Robinson)
acf5bc8 transaction: use better error typing for pegin destructuring (Philip Robinson)
6659691 blind: use array splitting in TxOut::unblind (fix potential DoS?) (Philip Robinson)
6005720 rewrite Address::from_base58 to eliminate all the unwraps (Andrew Poelstra)
d216a7c Add arrays.rs and slices.rs from bitcoin-internals 0.5 (Philip Robinson)

Pull request description:

  There are a number of useful improvements from #279 we would like to include in 0.26.x.

ACKs for top commit:
  apoelstra:
    ACK 6fee387; successfully ran local tests

Tree-SHA512: dd5d142dd76cbe472c67823989f79ed1bb1024282b788cd840bd4cf71db95ef277aa4eeb416ebe829a01f4dd22ed672dbc293e181bddaa9ccb0814e47dd1cc19