Removing FileTokenCache.add overload.

[marstr]

Fixes #26

[marstr]

Pro: Tests-pass
Con: The lock is now scoped so that it isn't for the duration of this library's add operation.

I'm not sure how much of a con that really is, as at the end of the day the only trouble would be with cache merging semantics. I'm open to your interpretation of that!

[rayluo]

Thanks for identifying such a tricky issue! This quick fix looks reasonable, so I'm giving an approval right now, to allow a quick subsequent release if needed.

More thoughts:

• Upstream repo does not have the same issue, because a re-entrant in-memory lock was used. In hindsight, this Extensions repo can potentially use a re-entrant lock provided by portalocker itself, but that seems like an undocumented feature, though.
• We would definitely want to have a unit test case coverage on this, in near future, if not immediately in this PR. Currently the test cases in this repo are mostly skipped by default, due to the lack of testing accounts in existing CI. We should still be able to have some unit tests which does not rely on a real test account. Something like this.