Improve CUDA 13.1 compatibility with newer glibc

[remer]

Summary

Pass -U_GNU_SOURCE -D_DEFAULT_SOURCE to Krasis NVCC invocations on Linux. This avoids the rsqrt/rsqrtf exception-specification conflict between newer glibc headers and CUDA 13.0/13.1 headers while retaining the standard/default interfaces.

Krasis already centralizes optional NVCC host arguments in nvcc_host_compiler_args(), so the compatibility flags apply consistently without changing the system compiler or non-Linux builds.

Observed failure

With CUDA 13.1 and glibc 2.43, CUDA kernel compilation fails with errors equivalent to:

bits/mathcalls.h: error: exception specification is incompatible with that of previous function "rsqrt"
crt/math_functions.h: previous declaration of "rsqrt"

Changing the selected host compiler did not resolve the collision because it is between the visible glibc and CUDA declarations.

Validation

• complete Krasis extension build succeeded after adding the flags
• all prefill, decode, HQQ, and supporting CUDA/PTX compilation completed
• the system's default compiler was not changed
• git diff --check

Compatibility scope

The observed Ubuntu 26.04/CUDA 13.1 combination is outside CUDA 13.1's validated distribution matrix, so this is a portability improvement rather than a claim that Krasis is defective on a supported CUDA configuration. CUDA 13.3 validates Ubuntu 26.04.

An open PR against NVIDIA's cuda-samples repository documents the same CUDA 13.1/glibc conflict and feature-macro workaround: NVIDIA/cuda-samples#403