✨ Relax condition on modifiers

[denialhaag]

Description

Currently, the modifiers in QC and QCO enforce that their bodies contain a single unitary operation followed by a yield operation. This has led to more complications than benefits. The condition should be dropped altogether, and arbitrarily many operations (unitary as well as, e.g., arith.constants), should be legal.

Acceptance criteria

• The condition is dropped.
• All relevant canonicalization patterns are updated accordingly.
• The translations from qc::QuantumComputation and OpenQASM to QC are updated accordingly. In particular, qc::CompoundOperations inside modifiers can be translated, and the test case mentioned in ✨ Translate qc::IfElseOperation to QC #1717 (comment) is added.
• Tests are added to cover new edge cases.

[burgholzer]

Adding to the list from above:

• The methods for computing unitary matrices are updated accordingly.
• Potentially add new canonicalizations, e.g., for pulling out global phases from modifier ops.

For the canonicalizations, a moderate change would be to limit them to the cases where only a single unitary is present. I think the resolution of inv or pow modifiers containing multiple operations should be a pass that is explicitly invoked when needed as this kind of simplification might not always be desirable.
Similarly the combination of adjacent operations with overlapping controls as well as splitting a CtrlOp with multiple ops into multiple CtrlOps might not always be desirable and one might need one or the other at different stages of the compiler pipeline.

[MatthiasReumann]

Some more thoughts:

• Operations such as arith.constant can be hoisted outside the ctrl-body for CSE.
• As discussed in Add RemoveSubsequentEquivalentCtrl Canonicalization Pattern #1686, the already existing canonicalization patterns for one-qubit unitaries should be able to create empty qco.ctrl operations, which are subsequently removed.