`struct` based encoding for metadata

[benjeffery]

The json metadata encoding in #491 can be inefficient for larger scale work. It should be possible to use struct to encode the metadata with a schema that looks like:

{
    "type": "object",
    "properties": {
        "accession_number": {"type": "number", "format":"L", "index":0}, 
        "collection_date": {
            "type": "string", 
            "format": "10s",
            "index":1,
            "pattern": "^([1-9][0-9]{3})-(1[0-2]|0[1-9])-(3[01]|0[1-9]|[12][0-9])?$"
        },
    },
}

Note the additional format and index keys that specify how to encode the metadata. The code would traverse the schema constructing a format string for struct using index as JSON objects are unordered. Note that we should stick to little-endian rather than letting struct decide.

[jeromekelleher]

Sounds good. I'm not convinced about order - wouldn't index be more natural (since we're counting from zero)?

[benjeffery]

I don't have a strong opinion, so index is fine.

[benjeffery]

format is already a JSONSchema keyword, so we'll need to use structFormat or similar. Details of how to extend the current validator are at https://github.com/Julian/jsonschema/blob/master/jsonschema/validators.py#L414

[jeromekelleher]

How about codec? Sorry, ignore! Need more coffee before making comments!

[benjeffery]

Closing this as #601 added the codec and we filed issues for further work (#632, #631, #630, #614)