fix(kompile): map linked NormalSym functions to monoItemFn(noBody)

[Stevengre]

Summary

• Generate MonoItemKind::MonoItemFn(symbol(...), defId(...), noBody) for linked NormalSym entries that are present in function_symbols but missing from items.
• Keep function resolution on concrete linked symbols instead of falling back to "** UNKNOWN FUNCTION **" for bodyless linked functions.
• Refresh affected integration/show fixtures and normalize panic symbol hash suffixes in fixture comparison to keep snapshot output stable.
• Preserve readable kmir show --statistics --leaves annotations for noBody callees (>> function, >> call span, >> message) even after replacing "** UNKNOWN FUNCTION **" with concrete symbols.

Context

In multi-crate SMIR, linked external functions can exist in function_symbols but not in local items (no local body). Before this change, that gap could leave lookupFunction(Ty) without a generated monoItemFn equation.

Red vs Green

RED (actual stuck shape in show fixtures):

#setUpCalleeData(
  monoItemFn(... name: symbol("** UNKNOWN FUNCTION **"), id: defId(38), body: noBody),
  ...
)

This is the unresolved-function path.

GREEN (after fix):

#setUpCalleeData(
  monoItemFn(... name: symbol("_ZN4core9panicking5panic17h<hash>E"), id: defId(38), body: noBody),
  ...
)

The same path now resolves to a concrete linked symbol instead of "** UNKNOWN FUNCTION **".

[Stevengre]

I think that it should be good to avoid UNKOWN FUNCTION in the semantics. Therefore, I bring this PR.

[dkcumming]

I know I have gotten to this late, but I want to bring up something related. We are not being consistent between instrinsics and other functions. For IntrinsicSym we demangle the name in the name field of the output of stable-mir-json, but for NormalSym we use the mangled name. @ehildenb had this PR converting to using the mangled name, I am not sure why the choice was to use the mangled name over the short name was used but I am going to guess it is because it is ambiguous which black_box is being called.

^^^ Notice black_box and assert_failed ^^^

Ideally I would like use a demangled but qualified name for everything. e.g.:

• _ZN4core9panicking19assert_failed_inner17h1d286061ca0adfe7E -> core::panicking::assert_failed_inner
• black_box -> core::intrinsics::black_box or core::hint::black_box

This would remove ambiguity and stop the need to do regex stuff to strip hashes etc.