[MilosLord]

ImInvSqrt currently uses _mm_rsqrt_ss even when IMGUI_DISABLE_SSE is defined. Its approximate result can differ between Intel and AMD CPUs, causing small differences that break pixel-exact render tests.

This change respects the flag by using the existing 1.0f / sqrtf(x) fallback, matching ImGui. Default behavior is unchanged.

clang++ -std=c++17 -O2 -march=x86-64-v3 [-DIMGUI_DISABLE_SSE] -c implot_items.cpp
llvm-objdump -d implot_items.o | grep -c rsqrtss

master, default:                      524
master, -DIMGUI_DISABLE_SSE:          524
this branch, default:                 524
this branch, -DIMGUI_DISABLE_SSE:     0