Set separate coordinate buffers to write queries

[stavrospapadopoulos]

Towards addressing #93

[Shelnutt2]

Serialization changes look good.

[joe-maley]

Do we need to worry about serialization compatibility with the REST server? For instance, if the REST server returns a query serialized with an older version, would we be able to properly deserialize it?

[joe-maley]

const-ref name

[joe-maley]

const size_t ret_size = attr_buffers_.size() + (coords_buffer_ ? 1 : coord_buffers_.size())
ret.reserve(ret_size)

[Shelnutt2]

Do we need to worry about serialization compatibility with the REST server? For instance, if the REST server returns a query serialized with an older version, would we be able to properly deserialize it?

The serialization format is buffer agnostic, previously we were using the list of attributes + coords these changes just rework things so instead of loop over attributes it loops over the list of buffers. For serialization it loops over the query buffer list, for deserialization we loop over the buffers as listed in the copy_state which comes from the attributeBufferHeaders object in capnproto spec. attributeBufferHeaders is an incorrect name, it really should be bufferHeaders but changing the name would break compatibility so it is left.

[stavrospapadopoulos]

@joe-maley the serialization format did not change in this PR. At a high level, the query serializes a set of named buffers. Supporting separate coordinate buffers really means allowing the named buffers to carry dimension names in addition to attributes and TILEDB_COORDS. This PR lifts some checks that mandated that the named buffers always correspond to attributes and TILEDB_COORDS.

[joe-maley]

Thanks! Just checking. LGTM.