[CIR][Dialect][NFC] Fix double whitespaces in cir.func assembly

[seven-mile]

This PR fixes the notorious double whitespaces introduced by visibility attribute, for cir.func only.

It uses "leading whitespace" for every print. And the printing of visibility attr is properly guarded by a check of !isDefault().

Double whitespaces in test files are removed.

[seven-mile]

Note that cir.global still has a similar problem, but for a little different reason.

cir.global uses MLIR assembly format with custom parser and printer custom<OmitDefaultVisibility>($global_visibility). When global_visibility has the enum case Default, it should not be printed. Thus, the custom printer function does nothing in this case. But an extra whitespace is still automatically emitted by MLIR assembly format generated code, resulting in double whitespaces.

Usually the best practice is to use the combination of I32EnumAttr and DefaultValuedAttr<>, which is recognized by assembly format and ensures the attribute is not printed for its default value (ref). Then, we can directly use DefaultValuedAttr<GlobalLinkageKind, "GlobalLinkageKind::ExternalLinkage"> and everything works.

But currently global_visibility in cir.global has type VisibilityAttr, which is a shallow wrapper of the underlying I32EnumAttr. This wrapper makes it hard to provide such a default value: we need a runtime MLIR context to build a default value.

Two thoughts on it:

• Replace VisibilityAttr with VisibilityKind, which is I32EnumAttr, or
• Use a null attribute as the default value of VisibilityAttr
  • Not ideal semantically, because default visibility != absence of visibility.

[bcardosolopes]

@seven-mile thanks a bunch for working on this, I filled #1029 with more details on what was the plan here.

Two thoughts on it

Which one do you think help us get to #1029 faster?