[QUANTITY] doc consistency

[Jonathan McDowell]

Pierre
 Thanks for your initial and very helpful editorial comments on Q.
 I have tried to address some of them in a slightly
 different new version V0.22. (mainly changed in sections 5 and 6.2.
 It's just a beginning, and I haven't dealt with the serialization questions,
 but let me know if it is the right direction.

 http://hea-www.harvard.edu/~jcm/www/vo/docs/qty22.pdf

 
 One general point on optional things: What do we MEAN by "optional"?

  - for each "get" method giving data you can extract from an object,
    (I avoid the words 'for each attribute' since we are talking interfaces)
    there may or may not be a special `null' return value (empty list, unitless tag,
    NaN value, etc.

  - for a given serialization, there may be default values for some
    tags, so that if that tag is omitted from the serialization, 
    the corresponding behaviour of the object is as if the default
    value had been given. Note that in principle different
    serializations don't have to allow the same things to be defaulted.

  - for a given constructor method for the object, the same is
    true: some things may have default values - but a different
    constructor method might have different defaults.

 I like the idea of most things being optional (in the sense of having
 defaults) in the *serializations*. 

 But when you distinguish between "UCD - Optional" and "Units - required, but
 default is <unitless>", I don't think that is a meaningful distinction.
 I don't think it makes sense to talk about things being default
 in the *objects* or the *interface*, and you have to be very careful
 when talking about things being 'optional'. Instead, it makes sense to
 talk about whether there is a concept of a null or special value
 that means 'information not available'.

 Everything is "required" in the special sense that you can always ask
 the question "what is your UCD?" "what is your unit?", "what is  your
 accuracy?", but might get the answer "I don't have one" (ucd-less,
 unit-less, empty accuracy list). 

 Many things are optional in the sense that if absent in the
 serialization, their value is a special value indicating  the absence of
 valid data. 

 Then there are a few things that it makes sense to provide non-trivial
 defaults for in the serialization (e.g. data type = string). But
 the concept of a default data type doesn't make sense in the object,
 just in the serialization and the constructors.
  
 So when you say "UCD - Optional" and "Units - required, but
 default is <unitless>", you actually mean:

 UCD:  getUCDString() may return blank or null string to indicate
 no associated UCD. This is the default if the tag is omitted on
 serialization.

 Units: getUnit() may return the special Unit value <unitless>.
 This is the default if the tag is omitted or given a blank value
 on serialization.

 Do you agree?


 - Jonathan

PS Apologies to Brian, you are quite right that I had earlier
conceded BQ has accuracy, I had forgotten that and fell into
the trap I've been angry at other people for doing, namely
going by my own instinct for truth and beauty instead of
checking the doc to remind myself what I agreed on.