Q4 cleanup

[sw]

Cleanup following #951 and #1046:

• remove unused ggml_vec_dot_q4_0
• warn for unused C functions (I didn't touch C++)
• use ggml_is_quantized for the work buffer calculation

[dfyz]

I think that everything inside this #ifdef should be removed as well. More precisely, bytes_from_q4_0_twoblocks_avx512() and dot_q4_0_twoblocks_avx512() should be removed, since they are only used it ggml_vec_dot_q4_0().

I'm guessing the CI didn't catch it with -Wno-unused-function because we only test AVX-512 under MSVC? Might be a good idea to include an AVX-512 build for Linux as well, instead of ACCELERATE (which is a no-op on Linux).

I just realized that CI wouldn't fail in any case because -Wno-unused-function is only a warning, not an error.

[sw]

You are right. Parts of this clever code may be useful for other quantization types, but that's what the git history is for.

We might want to add -Werror if CI should catch warnings.

[dfyz]

I don't know if I can/should approve this, but this PR looks pretty uncontroversial to me.