Update sortedmult, pkh, wpkh to the new validation framework. These
constructors now return a ValidationError rather than an Error, which
is a more precise type. The constructors also actually do validation;
in particular sortedmulti now does a duplicate pubkey check where it
didn't use to.

The theme over these next several commits is going to be "we now do
checks where we didn't use to". It may be that we get pushback from
users over this. In that case we need to adapt somehow but at least
we can do it in a principled way, rather than the current "there are
a ton of different ways to check and they're all randomly half
implemented".

This grammar is weird and hard to understand.

This is another big commit that has a bunch of behavior changes of the
form "now we enforce sanity rules where we didn't use to". This one lets
us delete a huge chunk of context.rs which is nice.

This eliminates the NonStandardBareScript error variant, but no tests
fail and I didn't replace the functionality. If I open a PR with this
commit in this state, we need to file an issue to restore this and to
add tests for it.

This is superceded by the validation error.

For some reason triggers rustfmt to reflow the MaxRecursiveDepthExceeded
variant. That will also be deleted in a later commit.

This is a big commit but it's almost all red.

This results in a bunch of error messages changing since we now fail due
to ValidationParams rather than due to Context::check_*.

One observable performance change is that we parse entire Miniscripts
before doing bounds checks. (This means that the error messages for
exceeded limits are now more accurate, though they will probably revert
when we do some optimization passes.)

There is a max recursion check in Expression::from_str and another one
in Miniscript::from_ast that I have left alone, for now, because of
this. The recursion check needs to fire immediately before constructing
any object that exceeds the maximum depth.

Later we will add a call to validate_non_top_level to Miniscript::from_ast.
This will let us drop the maximum depth check, but it will require an API
change because the caller will need to provide validation parameters.
So we leave it be for now.