Refactor state preparation routines

[tanujkhattar]

• [ ] StatePreparationAliasSampling uses QROM instead of SelectSwapQROM. We should make this a configurable choice.
• [ ] StatePreparationViaRotations prepares amplitudes using self._prepare_amplitudes(bb, **soqs) and phases using self._prepare_phases. The logic in self._prepare_phases can be pulled out in a separate Bloq.
• [ ] StatePreparationAliasSampling expects positive unnormalized_probabilities instead of amplitudes. We can update it to expect amplitudes and use strategy used by StatePreparationViaRotations; i.e. we first prepare the absolute value of amplitudes using alias sampling then delegate to the prepare_phases bloq above to prepare the right phases.

Both the state preparation routines are bottlenecked by QROM, so it'll be nice to have a way to configure which QROM bloq should be used to do the data loading (QROM vs SelectSwapQROM using clean or dirty qubits)

[fdmalone]

Re StatePreparationAliasSampling isn't the usual trick to take the absolute value, load signs in qrom and then apply phases using a CZ?

[tanujkhattar]

The amplitudes can also be complex in general -- suppose you want to prepare a state

$$\sum_{x} ||f(x)|| e^{i\phi(x)} \ket{x}\ket{\text{garbage}}_x$$

but state prep alias sampling gives you

$$\sum_{x} ||f(x)|| \ket{x}\ket{\text{garbage}}_x$$

You can then do qrom with addition into phase gradient to synthesize the complex phases $e^{i\phi(x)}$ (or club it with the QROM of alias sampling followed by addition into phase gradient, which is equivalent to what you are saying but for the case when phases are just +1 / -1)