Revisit decision to make `keyPosition` mandatory for dimensions

[hoehrmann]

In #46 the TWG decided to make keyPosition mandatory. I propose to remove or deprecate the property, or at least make it optional and advisory-only.

• The keyPosition property is not used in the interpretation of SDMX-JSON data messages. The association with the underlying data structure definition is by IDs, not by position. In SDMX-JSON structure messages the corresponding position property equally serves no purpose, since the order is given by index in the relevant array.
• It makes it look as though the order of dimension indices in colon-separated dimension index strings is related to the keyPosition.
• An SDMX implementation that receives data in SDMX-JSON form, what should it do when the data message is fine, except that some keyLocation property does not match with the DSD? That seems an unncessary error case.
• The SDMX-REST API no longer requires knowing the order of the dimensions in the DSD if you know their IDs (c parameter).
• In the issue above, loading the DSD for this information was considered burdensome because of limitations of SDMX implementations from that era, where multiple versions of SDMX-ML had to be supported for some data sources. Now, many years later, these implementations now also support SDMX-JSON structure messages, rendering the argument obsolete.

(In doubt, please handle this as a public review comment on SDMX 3.1 once the comment period begins.)