This commit is the first half and the meat of the implementation
of bytecodealliance#1774 where, by default, WIT processing tools will now merge world
imports when possible based on semver versions. The precise shape of how
this is done has gone through a few iterations and this is the end
result I've ended up settling on. This should handle all the various
cases we're interested in and continue to produce valid worlds within a
`Resolve` (with the help of `elaborate_world` added in bytecodealliance#1800). CLI
tooling has been updated with flags to configure this behavior but the
behavior is now enabled by default.

This commit implements the final bit of bytecodealliance#1774 where the `wit-component`
crate will now match imports based on semver version in addition to the
previous exact-name matching that was performed previously.

This is now possible with import version matching so refactor the
internal data structures to better support insertion of possibly
duplicate shims. Most of wit-component was already ready for this
refactoring, it was just the initial generation of shims that needed to
be reorganized slightly.

Any interface could be modified, so elaborate all of them to fixup anything
that needs new imports.

This commit updates how the semver-merging bits of `Resolve` work by
moving it towards the end of the encoding process rather than
once-per-world-merged. That helps both deduplicate work and fix some
asserts I'm seeing that are tripping if a `Resolve` is import-merged and
then merged again elsewhere.

This additionally simplifies some APIs because the boolean of what to do
isn't threaded to quite so many locations.

Dependencies might depend on replaced interfaces so the dependency edges
there need to be updated in the same manner as imports.

Also ensure to at least try to test the new merging function.