llama.cpp b10955 fixes heap corruption caused by a precompiled header disagreeing with itself.

The bug: the ggml-cpu PCH force-included ggml-impl.h before ops.h, pulling in <new> and defining __cpp_lib_hardware_interference_size, so C++ kernels used CACHE_LINE_SIZE 256 while the C work-buffer sizing in ggml-cpu.c used the fallback 64. The rope work buffer came up short by (CACHE_LINE_SIZE/4 - 16) * n_threads * 4 bytes and corrupted the heap, crashing later in an unrelated function. The fix disables the PCH and removes the std::hardware_destructive_interference_size branch entirely. A constant that changes value based on include order is a category of bug I'd never have found.