talk: Improvements in voicing "years"
* Enhance LANG_VOICED_DATE_FORMAT to distinguish between "numeric year"
and "grouped year" (2020 -> "two thousand twenty" vs "twenty twenty",
respectively)
* Metadata year voicing will now use "numeric year" if specified in
LANG_VOICED_DATE_FORMAT instead of always using grouped year.
* Datetime year voicing respects the format instead of always using
"numeric year"
Change-Id: Icc25da7c36107d3e4e8c70291f87a915e2bcabd3