refactor(circuit)!: build UCC from an ffsim UCCSD operator

[mrossinek]

Mirrors the UCJ change: the gate reimplemented the amplitude conventions and the parameter-vector packing alongside the part that is actually this package's concern, namely turning the amplitudes into a mapper-agnostic fermionic circuit. The first half duplicated ffsim, so this narrows the gate to the second half and takes the ffsim operator directly.

Verified rather than assumed: the removed num_parameters/from_parameters agreed with ffsim's n_params/from_parameters (14 parameters at norb=4, nocc=2, with identical t1/t2), and a restricted operator applied through this gate reproduces ffsim's own state vector exactly on coupled-cluster amplitudes.

Unlike UCJ, this drops capability ffsim cannot currently replace, which is why it is a separate commit: from_t_amplitudes, the spinless variant and the opt-in antisymmetric parameterization all go. Users needing those can build an Evolution over their own cluster operator, which also hands them control over the Trotter ordering of its terms. Note that ffsim ships no Qiskit gate for UCCSD at all, so this gate remains the only route from one of its UCCSD operators to a circuit -- under Jordan-Wigner or any other encoding.

What remains is the part with no ffsim equivalent: the cluster generator and its conjugate-paired grouping, which keeps every factor of the product formula Hermitian and hence unitary. A term-by-term split would not even preserve the norm, so the grouping tests are kept verbatim; the two that previously reached this machinery only through the spinless variant are ported to the restricted one, which exercises the same paths on a spinful register.

One pre-existing convention difference is now documented rather than changed: the cluster operator only ever sees the part of a same-spin t2 that is symmetric under the simultaneous exchange t2[i,j,a,b] = t2[j,i,b,a], while ffsim reads the raw tensor. Coupled-cluster amplitudes always carry that symmetry (verified against PySCF), so the two agree exactly on any physical input; only a hand-built asymmetric tensor can tell them apart.

Refs #318

[coveralls]

Coverage is 98.165% — ucc-from-ffsim-op into ucj-from-ffsim-op. No base build found for ucj-from-ffsim-op.