Bootstrap loop iter args (Halo Solution A-2)

[j2kun]

From 4.2 of https://dl.acm.org/doi/10.1145/3669940.3707275

Solution A-2: Bootstrapping all the loop-carried variables at the beginning of each iteration. Regardless of consumed levels of ciphertexts, bootstrapping reinitializes their level. Therefore, inserting the bootstrap operation at the beginning of the loop iteration can resolve the level type mismatch problem. Figure 2: Solution A-2 depicts the level matching process of the input and output of the loop body. Before entering the loop body, HALO reduces each level of y and a to the minimum level using modswitch for live-in and loop-carried variables. Then, HALO recovers the levels of the loop inputs to 10 by bootstrapping all loop inputs at the beginning of the loop. After type matching process, the loop is transformed to a type-matched loop (repeatable). However, the type-matched loops are inefficient. For efficient loop execution, bootstrap-based loop restructuring is required.

For this PR I introduced a new mgmt op: mgmt.level_reduce_min which corresponds to "reducing the level to the minimum allowable level." I suspect we need this op in this situation because we will be inserting this level_reduce_min before knowing what the actual level budget is, and this ensures that our level analysis can reach a fixpoint when processing the loops that will be present after the HALO implementation is integrated.

Rebased over #2534, and similar to that PR, I just created a pattern to handle the core operation, wrapped it in a dedicated pass for testing, and will figure out how to integrate it into the mgmt pipeline once all the remaining pieces of the Halo paper are implemented.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[j2kun]

The CI failure was due to llvm/llvm-project#175056 (comment) which will be fixed in the next LLVM integrate.