Remove preliminary late includes cy28

[cxzhong]

The 'preliminary_late_includes_cy28' directive was a temporary workaround introduced in Cython 0.28 (GH#2079) to control include ordering. Now that macros.h is included before Cython emits its own declarations, we fix this properly by:

• Add extern forward declarations for cysigs and helper functions directly in macros.h, making it self-contained
• Add an unlikely() fallback macro in macros.h for early-include contexts
• Remove static from exported symbols in implementation.c and add CYSIGNALS_API visibility attribute so they are accessible from other extension modules
• Move shared symbol declarations in signals.pxd from the Cython cdef nogil capsule mechanism to cdef extern from macros.h
• Link tests extension against signals in meson.build