lattigo: lattigo.bootstrap needs ResetsLevelOpInterface

[AlexanderViand]

Driven by #3225, where I ran into

--- FAIL: TestLoLA (20.83s)
    lola_test.go:65: Lola__preprocessing took 522.491416ms
panic: cannot Rescale: input Ciphertext level is too low [recovered, repanicked]

The bug had a really simple fix (lattigo.bootstrap seems to have lost its level-analysis annotation when we switched from ResetsMulDepthOpInterface to ResetsLevelOpInterface) but really annoying to track down. As a result of the missing level reset annotation, AllocToInplace thought that a mul op after a bootstrap was still happening at basically level 0, so tried to re-use level-0 buffers for them. Lattigo, in its wisdom, would not throw an error upon using a too-small/too-low-level output buffer, but instead just force things down to level 0 and hence in the next step, we'd get an attempt to rescale a level 0 ciphertext 🤦

[asraa]

while i know this is right (hence lgtm).. it broke my compilation of lenet when i patched it :|

[j2kun]

It looks like this PR still has //third_party/heir/tests/Dialect/Lattigo/Transforms:alloc_to_in_place_levels.mlir.test failing

[AlexanderViand]

It looks like this PR still has //third_party/heir/tests/Dialect/Lattigo/Transforms:alloc_to_in_place_levels.mlir.test failing

Oh, good catch, I had a simple error in the FileCheck, but now I'm very confused why the CI passed?

EDIT: is it possible that with pull_request_target, github.sha points to main, rather than the PR?

[asraa]

EDIT: is it possible that with pull_request_target, github.sha points to main, rather than the PR?

I noticed the same thing with CI :|