Add type inference for LWE/CKKS ops

[j2kun]

This came up in the conversion from Orion to CKKS (#2326). That line of work requires me to reimplement chebyshev evaluation and the halevi-shoup kernel, but at the level of CKKS with parameters pre-selected.

#2338 is addressing the finer points of that conversion, but in the mean time it has proven helpful to extend the result type inference capabilities of the LWE/CKKS dialects, because when I insert ciphertext mgmt ops at the CKKS level, I need to re-walk the IR and re-infer result types for all ops.

Yes, I know this is what @ZenithalHourlyRate warned against and why we have mgmt at the higher level. Normally I wouldn't want to do this either, but for comparing against another compiler that handles ciphertext management and parameter selection itself, I don't see another way :)

[j2kun]

For the reviewer: I'm looking for some extra scrutiny on the logic of the type inference, to make sure I'm not incorrectly dropping limbs or dividing by the wrong moduli.

[ZenithalHourlyRate]

LGTM!