`float16` and `bfloat16` support

[asg017]

create virtual table vec_movies using vec0(
  synopsis_embedding float16[768]
);

create virtual table vec_movies using vec0(
  synopsis_embedding bfloat16[768]
);

Also vec_quantize_float16() / vec_quantize_bfloat16()?

[maruel]

bfloat16 would be awesome for storage efficiency.