test: document whole-function assembly kernels (#28447)

[quaesitor-scientiam]

Part 3 of #28447: whole-function / naked assembly support.

Summary

• Document the external .S plus .o linkage path for kernels that need their own frame or
  call instructions.
• Add a whole-function SHA-256 compression kernel in amd64 assembly.
• Call the kernel from pure V through an explicit @[c_extern] C ABI declaration and verify the
  SHA-256 abc vector.

Validation

• GCC full assembly suite: 8 passed, 10 skipped.
• GCC external-kernel fixture: passed on Win64.
• Clang external-kernel fixture: passed on Win64.
• SysV amd64 form: assembled with Clang Linux target.
• MASM ml64 object plus MSVC pure-V caller: passed locally.
• Repository MSVC fixture: skipped because the checked-in source is GNU .S; MSVC users can
  provide a MASM-compatible .obj as documented.
• TCC: fixture skipped because its assembler does not support the .macro directives used by
  this kernel.
• check-md doc/docs.md: 300 OK, 0 warnings, 3 existing platform/environment errors (dprintf,
  SQLite headers, and malformed ml fence) plus 1 existing formatting error.