This commit ports the consolidate blocks pass to rust. The logic remains
the same and this is just a straight porting. One optimization is that
to remove the amount of python processing the Collect2qBlocks pass is no
longer run as part of the preset pass managers and this is called
directly in rust. This speeds up the pass because it avoids 3 crossing
of the language boundary and also the intermediate creation of DAGNode
python objects. The pass still supports running with Collect2qBlocks for
backwards compatibility and will skip running the pass equivalent
internally the field is present in the property set.

There are potential improvements that can be investigated here such as
avoiding in place dag contraction and moving to rebuilding the dag
iteratively. Also changing the logic around estimated error (see Qiskit#11659)
to be more robust. But these can be left for follow up PRs as they
change the logic.

Realistically we should look at combining ConsolidateBlocks for it's
current two usages with Split2qUnitaries and UnitarySynthesis into
those passes for more efficiency. We can improve the performance and
logic as part of that refactor. See Qiskit#12007 for more details on this
for UnitarySynthesis.

Closes Qiskit#12250

The test failure fixed by a test change was incorrect and masked a logic
bug that was fixed in a subsequent commit. This commit reverts that
change to the test and removes the release note attempting to document a
fix for a bug that only existed during development of this PR.

This commit reworks the logic to reduce the number of Kronecker products
and 2q matrix multiplications we do as part of computing the unitary of
the block. It now computes the 1q components individually with 1q matrix
multiplications and only calls kron() and a 2q matmul when a 2q gate is
encountered. This reduces the number of more expensive operations we
need to perform and replaces them with a much faster 1q matmul.

Co-authored-by: Kevin Hartman <[email protected]>

Co-authored-by: Henry Zou <[email protected]>