checksum module needs some love

[tcharding]

After working on #608 I noticed a few things:

• We seem to be using fmt::Write::write_str with the implied suggestion that we avoid allocation of a string but I think that fmt::Argumenst::as_str() gets called and only avoids allocation for static strings (no uses of placeholders {}), since we always use {} I'm reasonably confident that this is allocating.
• There are some optimisation possibilities by using iterator adapters instead of strings, ie take an iterarator of chars and yield the chars + # + the checksum.
• The public API is a bit messy:
  • checksum::Formatter should not be public because it is not really a general purpose formatter, it is specifically for writing a descriptor to a formatter and overriding the use of alternate to control the addition of the checksum.
  • descriptor::checksum::desc_checksum could be better named as could descriptor::checksum::verify_checksum to use Rust naming convention (ie., users should be using a single path on functions so checksum::verify seems better, not sure about the other one.
• Error handling could be improved if/when we start to do errors elsewhere

Related:

• Descriptor loses formatting options #323
• String descriptors can panic on to_string #554

[apoelstra]

Yeah, I agree with all these things. Confirmed that elements-miniscript is not using the checksum::Formatter type (it has its own).