[CIR] Track size_t and int size with module attributes

[AdUhTkJm]

To give LoweringPrepare type information from CIRGenTypeCache, this PR adds two attributes to ModuleOp:

module attributes {
  cir.int_size = #cir.int_size<32>,
  cir.size_type_size = #cir.size_type_size<64>,
  ...
} {}

The CIRDataLayout class is also extended to have getPtrDiffTy and so on.

Some tests that only expects cir.lang and cir.sob are also changed to take this into account.

[Lancern]

A small question here: would it be better to provide all these type-layout related information within a single attribute, instead of spreading them into separate attributes?

For example, would it be better if we update this PR to instead provide a single attribute like this:

module attributes {
  cir.type_sizes = #cir.type_sizes<
    int = 32,
    size_t = 64,
  >,
} {}

[AdUhTkJm]

A small question here: would it be better to provide all these type-layout related information within a single attribute, instead of spreading them into separate attributes?

For example, would it be better if we update this PR to instead provide a single attribute like this:

module attributes {
  cir.type_sizes = #cir.type_sizes<
    int = 32,
    size_t = 64,
  >,
} {}

That makes sense! Now I just changed it to be exactly like this.

[bcardosolopes]

@AdUhTkJm thanks for working on this - really good point @Lancern. My suggestion is that we do this for all types in CIRGenTypeCache that don't have an explicit bit-width attached to the name (FloatTy, DoubleTy, UIntPtrTy, SizeTy, etc). Few extra things:

• We don't need to add an entry for all of them if unnecessary (e.g. unions in CIRGenTypeCache can be represented by the same entry but given two different getters in the attribute extraDeclarations).
• Make sure you initialize the attribute before CIRGenTypeCache and then initialize the relevant members using the information from the attribute.
  The long term plan here (not for this PR) would be to kill CIRGenTypeCache in favor of only using the attribute in both CIRGen and elsewhere, at which point we can have more getters in the attribute to also populated the more obvious types, without storing any information like SInt8Ty, etc.

[bcardosolopes]

Comment above

[bcardosolopes]

Great, one more round of review