use semver trick to replace most types with types from 0.26

In 0.26.0 we made two breaking changes: we changed the signature of
some blinding functions in PartiallySignedTransaction, and we removed
the & from a couple utility methods in pset::raw::ProprietaryKey and
in TxOut.

Therefore, those three types, and any type which refers to them in their
public API, have semver-breaking changes. But no other type does! So we
can re-export everything else in a minor release of 0.25.x.

The types that we must preserve are:

* PartiallySignedTransaction
* TxOut
* Transaction (contains Vec<TxOut> in a public field)
* Block (contains Vec<Transaction> in a public field)

There are then several modules which have impl blocks on these types,
which we need to keep; and the PSET `Output` type has accessors for
a `TxOut`, so we need to keep that, which means that we need to keep
its whole `pset::Serialize` impl, which means we need to keep the
whole private `Map` trait. Frustrating.

Now, this commit is very large and may be hard to review. However, much
of it is easy:

* First, we entirely delete the `address`, `blech32`, `confidential`,
  `dynafed`, `encode`, `error`, `ext`, `fast_merkle_root`, `hash_types`,
  `hex`, `issuance`, `locktime`, `opcodes`, `parse`, `schnorr`, `script`,
  and `taproot` modules and replace them with re-exports.

  To confirm these changes are OK, you ought to check their API on
  crates.io (or by just grepping) and see that none of the types listed
  above appear there.

* Then, the `block` and `transaction` modules have every type except for
  `TxOut`, `Transaction` and `Block` removed. So that's probably also
  easy to review, though again it's good to check that these types don't
  appear in the API of anything that was removed.

* Then `blind.rs` and `pset/elip100.rs` have impls on `Transaction` and
  `PartiallySignedTransaction`, so they had to stay.

* Then `sighash.rs` has some functions which are conditioned on T with
  T: Deref<Transaction> or Borrow<TxOut> so these had to be kept; we
  also had to keep the private method split_anyonecanpay_flag and move
  it to a standalone function.

* Finally, the whole PSET module was kept, except that I removed the
  Serialize trait and a couple utility structs, including the errors,
  in a somewhat ad-hoc way.

PROBABLY it's fine to just skim this and we'll release it and see if
anything breaks.

Author: apoelstra