Tag normalization: normalize abbreviation special cases

[Tiendil]

The general abbreviation processing will be implemented in gh-422

But there may stay some special cases, mostly from news native tags, that we should handle separately/differently.

Examples:

• u.s.a. -> usa
• u. s. a. -> usa
• u s a -> usa
• ph.d. -> phd (there a lot of abbreviations with multi-letter parts, like ph)
• but .net -> dot-net
• but r programming -> r-programming

We may need to add checks in multiple places:

• in tags.converters to detect some patterns in raw tags;
• in tags.normalizers to detect some patterns after initial normalization.

Attention: check the statistics of such tags in the DB before implementing anything, to avoid overengineering.

Internal task id: ff-524