Fix handling of trailing whitespaces in initials' delimiters

[PgBiel]

Fixes #137

Some styles, such as vancouver, require there to be nothing between initials, so Astley, Rick Gavin would be formatted as Astley, RG instead of Astley, R. G. (more usual) or Astley, R G for example. However, even when setting initializer-with="" (empty delimiter), Hayagriva was still forcing whitespaces between initials.

It was thought that #150 would have fixed it by only adding whitespaces if a delimiter was indeed specified, but it actually didn't change anything as there is always a delimiter - however, it is sometimes an empty string. While adding an explicit check for empty strings would have fixed this specific problem, it turns out Hayagriva's behavior of forcing a whitespace to always be added after the delimiter, replacing any of its trailing whitespaces if there are any (otherwise, one is added anyways), is not consistent with other CSL processors. Indeed, several citeproc tests demonstrated that the expected way to add whitespaces between initials is through initializer-with=". " instead of initializer-with=".". So, I removed Hayagriva's automatic whitespace behavior, except when we're replacing hyphens (e.g. for initializer-with=". ", Aa-Bb should become A.-B. and not A. -B.).

It is, however, unfortunate that this seems to be somewhat inconsistent with a specific example given by the relevant section of the CSL 1.0.2 spec:

`initialize-with`
    When set, given names are converted to initials. The attribute value is added after each initial (“.” results in “J. J. Doe”).

...but most official styles I've seen appear to assume that whitespaces are preserved, using ". " instead of "." in that case. So this seems to be the way to go.

I've also added some tests. Two more citeproc tests are now passing as well.

I'll be keeping this as a draft as I still need to test the case with initialize="false", initialize-with=".", but in principle the code seems to preserve whitespace here as well, however whitespaces from the name itself are also kept, which is why initialize-with="." is used instead of initialize-with=". " by official styles in this case.

[reknih]

If the other implementations and styles assume/exhibit/rely on this behaviour, we should also add it, so I'm good with it from that perspective.

[PgBiel]

Taking a look here, it seems that automatic whitespaces should also be removed when initialize="false" but initialize-with is set, according to citeproc tests, whereas we currently always add whitespaces. But I'll leave that to a separate PR.