Implement a baseline packing/layout conversion algorithm

[j2kun]

As we start to add support for different RLWE packings to HEIR, we will need to implement packing layout conversion operations and lowerings.

Once we have an attribute for layout (cf. https://github.com/google/heir/issues/913), and have decided on what types to store the attribute, we will need:

• A switch_packing op that changes the layout attribute. The simplest example of this is a single ciphertext whose data elements should be rearranged.
• A lowering that implements layout conversion via the simplest possible approach: select elements with the same shift, mask and rotate, and then sum up all shifts done this way. (Sec 3.1 of https://eprint.iacr.org/2014/106 describes this as the naive algorithm)

Other issues will handle optimization passes at the switch_packing op level to reduce the cost of switching packings (this is what Fhelipe does in #800), and more advanced lowerings that create deeper shift networks (larger multiplicative depth) to allow for a smaller number of rotation groups.

[AlexanderViand-Intel]

While this is the issue for the baseline approach, I wanted to throw in that Vos et al., Efficient Circuits for Permuting and Mapping Packed Values Across Leveled Homomorphic Ciphertext, ESORICS'22 claims to outperform the (non-naive) HElib approach significantly (of course, never trust a benchmark you haven't cherrypicked yourself... 😉)