Make strings for data usage easier to localize

The strings were complex to localize, especially since they used
formatting that needed to be carefully reconstructed for cases where
the amount of data appears in the middle of a string. Also, we
already had a localization for file sizes, which we should avoid
redoing, to minimize discrepancies.

Finally, we are now bidi wrapping the usage amount, as it may break
in the RTL locales otherwise, since the RTL locales may be using
Latin abbreviations for gigabyte etc that are LTR.

Since formatting needs to be applied to everything except the number
of units in the amount of data used, and markup should be avoided
since it will confuse bidi wrapping, the code carefully constructs
CharSequence templates that are then decorated with spans after
getting filled in.

Bug: 28747101
Change-Id: I18756b4de99eb551a51fe5e74d1c9cf505551f08