fix(typecheck): correct d: and hash-fragment witness counting

[guggero]

Two of the values that #859 (f57cac0 "typeck: pull satisfaction/
dissatisfaction limits into struct") moved into SatData ended up in the
wrong field, so ExtData no longer agrees with the witness the satisfier
produces:

• cast_dupif used to add (w + 2, s + 1) to the (witness size,
    scriptsig size) pair and 1 to the element count. It now adds 1 to the
    witness size and 2 to the element count: the witness size took the
    scriptsig's increment and the count took the witness size's. A d:
    satisfaction pushes a single element holding 0x01, which is two bytes
    in a witness (its length prefix and the byte itself) and one byte in a
    scriptsig (OP_1), one element either way. Placeholder::PushOne
    sizes that element as two, and or_i, whose <1> selector is the very
    same element, still adds (2, 1).

• The four hash fragments used to have a stack_elem_count_dissat of 1
    and now have a dissat_data.max_witness_stack_count of 2, which is the
    value of the exec_stack_elem_count_dissat that sat next to it.
    Dissatisfying a hash fragment pushes a single 32-byte element, as the
    33-byte max_witness_stack_size beside it already says.

The witness size of d: is the consequential one, because it is an
under-estimate and max_satisfaction_size feeds max_weight_to_satisfy
for every descriptor type: a bare, sh, wsh or tr descriptor whose
miniscript contains a d: wrapper under-states the weight of its input by
one weight unit per wrapper, so a fee computed from it lands below the
intended feerate. The two element counts are over-estimates, which only
cost their users precision.

Both are in 13.0.0, 13.0.1 and 13.1.0. This area of the same refactor has
needed one follow-up correction already (f3ea32c, the thresh execution
stack count).

Found while differential-testing an independent port of this crate, which
disagreed on the witness size of the expressions whose satisfaction goes
through a d: wrapper (242 of ~16k) and on the element count of the hash
dissatisfactions.