[MooreToCore] Alternative conditional lowering and type conversion fixes

[maerhart]

• Alternative conditional op lowering to a mux if there are no relevant side-effects in the regions
• Add missing type conversions (mostly target types not marked legal and target types with nested source types not supported). This is necessary to make the conditional lowering work because the dialect conversion framework pushes an InlineBlockRewrite to the queue with triggers a signature conversion call to the function body the conditional op is in, which failed because the target types were not marked legal.
• Use the type converters isLegal helpers to determine dynamic legality instead of a custom solution.

[fabianschuiki]

LGTM! The isLegal query is really neat and cleans stuff up very nicely.

[hailongSun2000]

I have a personal question regarding typeConverter.addConversion([&](hw::InOut/Array/StructType type).... If we convert moore.ref/array/struct to hw.inout/array/struct, would we get an error type like hw.inout<moore.i32> 🤔?

[maerhart]

I have a personal question regarding typeConverter.addConversion([&](hw::InOut/Array/StructType type).... If we convert moore.ref/array/struct to hw.inout/array/struct, would we get an error type like hw.inout<moore.i32> 🤔?

Not with this set of conversion callbacks, because we always convert the inner type as well (and return early if that was unsuccessful) and because we now also have conversion callbacks for the target types that also convert the inner types. Before this PR, if you have such a type in the IR for some reason, it would not be properly lowered.

[hailongSun2000]

I understand these callbacks but sorry for not realizing which special case will trigger this error 😅.