Control behavior of zeros in C level vec_as_location()

[DavisVaughan]

This would bubble up to R through num_as_location()

We need another option in vec_as_location_opts for controlling the behavior of zeros. I proposed this to @lionel- , who seemed to agree:

• remove: The current behavior, removes zeros. Problematic in some cases because this changes the size of the index.
• ignore: Leaves 0 values in place, unchanged.
• error: Errors with a custom classed error for this case