fix: return error when parsing xprv into DescriptorPublicKey

[trevarj]

Refactor FromStr for DescriptorPublicKey so that it can parse a xprv/tprv but
then error with a new variant DescriptorKeyParseError::UnexpectedXPrivateKey
so that the user knows to use
Descriptor<DescriptorPublicKey>::parse_descriptor() instead.

Added test that reflects what was reported in the issues:
bitcoinfuzz/bitcoinfuzz#70
#785

[apoelstra]

This feels a little dangerous to me. If the user provides secret data and then we irreversibly throw it away, could that cause a loss of the secret?

[trevarj]

we irreversibly throw it away

Yeah, that's a good point. Using parse_descriptor seems like the best route since you get back the KeyMap.
Wonder if FromStr should just be removed. I feel like most people will see parse_descriptor in the API before they consider calling parse(), or maybe that's just me.

[apoelstra]

Wonder if FromStr should just be removed. I feel like most people will see parse_descriptor in the API before they consider calling parse(), or maybe that's just me.

parse is a stdlib function that users will reach for before looking at our API for even a second.

And given that descriptors are, at their core, a string representation of an output, FromStr is a pretty fundamental trait.

Having said this, we may want to add a special error variant to FromStr for DescriptorPublicKey saying that it appears we're trying to parse an xprv, and maybe even suggesting to use parse_descriptor.

[apoelstra]

We can also add an allow_private_keys boolean to ValidationParams in #957, which is explicitly documented to convert private keys to public ones, and then people can use from_str_with_params to explicitly opt into this behavior.

That could be used for the fuzztests so that they would pass.

[trevarj]

That could be used for the fuzztests so that they would pass.

Why not just change the fuzz tests to use parse_descriptor? Then return a new error variant in FromStr with a warning to use parse_descriptor. No need for the extra from_str_with_params IMO

[apoelstra]

Yeah, good idea.

[apoelstra]

ACK 1d22086; successfully ran local tests