Fix: llvmcall segfault/parser error with zero-sized arguments

[JSmith2023]

The Base.llvmcall intrinsic previously crashed (Segmentation Fault) or threw a parser error when given a zero-sized type (ZST), such as nothing, as an argument.

This patch addresses two issues in src/ccall.cpp:

Adds a null check (if (!v)) in llvm_type_rewrite to safely handle ghost arguments.

Skips void types when generating the LLVM function signature string in emit_llvmcall, ensuring valid syntax.

The result is that calling llvmcall with a ZST now correctly throws a user error if the user attempts to reference the ZST argument in the custom assembly.

Fixes #60063

I have also included a regression test in test/llvmcall.jl to prevent this issue from recurring.

[JSmith2023]

Update on CI Failures:

I've pushed an amendment to address the failures seen in the linux-gnuassertrr and aarch64 builds.

Fixed Segfault in sysimage build: The strict assertrr build was crashing because nullptr was being passed to llvm_type_rewrite for ZST arguments. I've updated the loop in emit_llvmcall to explicitly skip the rewrite for Void types and assign a safe placeholder (NULL), which avoids the crash without creating illegal UndefValues for void types.

Fixed LLVM IR Syntax Error: The previous logic for skipping ZSTs in the function signature could generate invalid commas (e.g., leading/trailing commas) in some edge cases, which caused failures on AArch64. I replaced the boolean flag logic with a counter to ensure commas are only emitted between valid arguments.

[Zentrik]

You seem to have formatted the file which makes the diff rather large, I would suggest undoing the formatting changes.