Reduce binary size by removing unused functions even when not using LTO (~200-900kb reduction)

[dsnopek]

It appears that without LTO, we were only passing the flags for the compiler to remove unused compilation units, not unused functions.

This PR attempts to add those flags for both GCC and Clang (I tried a bunch of stuff with MSVC, but wasn't able to find anything that reduced size at all):

Platform / Compiler	Size Before	Size After	Change
Linux / GCC 13	2,866 KB	1,945 KB	-921 KB (-32.1%)
Linux / Clang 18	3,063 KB	2,443 KB	-620 KB (-20.2%)
Windows / MinGW GCC 13	2,100 KB	2,089 KB	-11 KB (-0.5%)
Android / NDK 28.1	1,161 KB	466 KB	-695 KB (-59.9%)
macOS / Clang 21 (w/ debug_symbols=yes)	1,045 KB	831 KB	-214 KB (-20.5%)

Re Windows: It seems MingGW can't fully do this optimization without LTO, which is why (a) the change is so small and (b) it doesn't add the -ffunction-sections -fdata-sections flags because that actually makes the size bigger, not smaller

Re macOS: We were already using -Wl,-dead_strip with debug_symbols=no, so this only adds it in the debug_symbols=yes case

With these changes, the effect of using either LTO or build_profile is much smaller, which is nice, because LTO is slow and using build_profile requires extra work