Improve ATTR_HTML keyword parsing

Org mode separates kvs not as initially assumed by whitespace (~ csv) but
rather at keywords (~ :\w+).

This is still not replicating Org mode behaviour though as I decided against
attributes ignoring multi-definitions. Instead we stack their
values (and those existing on the element) for certain attributes (class, style
for now).

e.g.
[[foo]]

would become <foo class="a"> in Org mode but becomes <foo class="a b"> with
go-org.

org/testdata/keywords.org

@@ -1,13 +1,14 @@
 
 #+CAPTION: and _multiple_
 #+CAPTION: lines of *captions*!
-#+ATTR_HTML: :class "a b c"
-#+ATTR_HTML: :id it
+#+ATTR_HTML: :class a b
+#+ATTR_HTML: :id it :class c d
 #+BEGIN_SRC sh
 echo "a bash source block with custom html attributes"
 #+END_SRC
 
 and an image with custom html attributes and a caption
 #+CAPTION: kittens!
-#+ATTR_HTML: :style "border: 10px solid black"
+#+ATTR_HTML: :style height: 100%; :id overwritten
+#+ATTR_HTML: :style border: 10px solid black; :id kittens
 [[https://placekitten.com/200/200#.png]]