balar: extract CUDA-independent packet header

[nab880]

Balar's packet definitions currently require CUDA headers and C++ namespaces, preventing standalone C and firmware peers from sharing them. Extract the definitions into balar_packet_wire.h, preserve existing CUDA-enabled C++ packet declarations, and keep balar_packet.h as the existing runtime-facing wrapper. Include the new header in the build and staged runtime tests.

This layer contains only the shared header, compatibility handling, build/staging changes, and focused C/C++ compilation and layout tests. Existing runtime sources are unchanged.

Validation:

• Installed SST test runner: 4 passed, 0 skipped, including C11/C++11 checks with NVIDIA CUDA 11.8 headers.
• Packet declarations compared against devel retain existing CUDA-enabled C++ semantics.
• git diff --check passed.
• The standalone fallback remains an opaque representation of CUDA-specific types; this split does not introduce a new versioned serialization format.

First layer of the balar-* replacement stack for #2690, targeting current upstream devel.

Stack (review order):

1. balar-1-wire-header
2. balar-2-test-support
3. balar-3-doorbell-driver
4. balar-4-compact-mmio