Add support for more metadata field types

[jan-niestadt]

Currently, BlackLab supports metadata fields with these types:

• tokenized or untokenized (text)
• numeric (integer)

It would be useful to include more types that Lucene supports natively, such as:

• specific numeric types: integer, long, float, double
• range of one of the number types, e.g. integer-range
• byte-vector / float-vector (nearest neighbour) (e.g. for embeddings)
• lat-long (latitude/longitude)
• etc., see Field

A date type would also be useful, although that doesn't seem to be natively supported by Lucene. If we add date fields, we should be very clear about what we do and do not support (only e.g. YYYY-MM-DD? Partial dates like YYYY or YYYY-MM or MM-DD? Or even a full timestamp with timezone etc. (probably don't do this until there's a clear need)

With these explicit field types, the frontend could choose a more appropriate default widget for the field.

(#618 describes a use case for floating point)

[jan-niestadt]

An added wrinkle is that span attributes would ideally behave exactly like document metadata, but for a part of the document.

In reality, this is not how it is. Span attributes cannot be tokenized, don't currently support numeric or other types, and the way they are indexed, it would be a challenge to achieve that.

[jan-niestadt]

See this Lucene issue for a discussion about vector-per-word. Looks like this doesn't address other types.