Add a name token table so on-demand typeaheads can match last names

Summary: See T585. We currently don't match middle/last/nth names in on-demand
tokenizers. Build a table so we can match them.

Test Plan:
Ran upgrade script, verified table looks sensible. Searched for "priestley" in a
tokenizer, got a bunch of test account hits.

mysql> select * from user_nametoken;
+-------------------+--------+
| token             | userID |
+-------------------+--------+
| evan              |      1 |
| priestley         |      1 |
| epriestley        |      1 |
| epriestley2       |      2 |
| ducks             |      4 |
| epriestley3       |      4 |
| asdf              |      6 |
| epriestley99      |      6 |
...

Reviewers: bh, nh, jungejason, tuomaspelkonen, aran

Reviewed By: aran

CC: aran

Differential Revision: 1034