BugFix in `SelectSwapQROM` when multiple datasets are specified

[tanujkhattar]

Two bugfixes:

1. Use self.use_dirty_ancilla in _value_equality_values_ -- This is important to make sure bloq.t_complexity() returns different values when bloq is initialized with use_dirty_ancilla=True vs use_dirty_ancilla=False. Right now, they can return the same value because the t-complexity is cached
2. qrom_batched_target[0] can return a numpy array because the target shapes can be multi-dimensional. This is especially true when input data is multi-dimensional (i.e. data[N][M] --> bached_qrom_data[N / k1][M / k2]). We need the target_0_0_..._0'th register and not the target_0th register; which can be multidimensional.

The added tests fail on main but pass with this PR.

[fdmalone]

We need the target_0_0_..._0'th register and not the target_0th register; which can be multidimensional.

I was a little confused by the underscores here given our weird target register convention, you just mean you want the actual zero'th target register not the zero'th row (in the 2D example)

[tanujkhattar]

I was a little confused by the underscores here given our weird target register convention, you just mean you want the actual zero'th target register not the zero'th row (in the 2D example)

Yup

[mpharrigan]

Why are we using value_equality_values at all?

[tanujkhattar]

Why are we using value_equality_values at all?

We store tuples of numpy arrays, which don't support value equality by default.

[fdmalone]

In several of the chemistry bloqs I ended up using tuples of tuples etc which is horrendous, but maybe this is solved in attrs now: https://www.attrs.org/en/stable/comparison.html#customization?

[tanujkhattar]

@fdmalone @mpharrigan Removed the use of _value_equality_values_ with the fancy attrs functionality Fionn mentioned above. This is ready to go now.