Very slow parsing of an invalid date

[lopuhin]

First several calls are above 1 second for me (python 3.6 on OS X):

In [1]: import dateparser

In [2]: %time dateparser.parse('not a date')
CPU times: user 1.94 s, sys: 99.2 ms, total: 2.04 s
Wall time: 2.13 s

In [3]: %time dateparser.parse('not a date')
CPU times: user 1.32 s, sys: 7.37 ms, total: 1.32 s
Wall time: 1.33 s

After some time it goes down to 150 -- 200 ms, which is still quite slow:

In [6]: %time dateparser.parse('not a date')
CPU times: user 145 ms, sys: 2.08 ms, total: 147 ms
Wall time: 152 ms

In [7]: %time dateparser.parse('not a date')
CPU times: user 147 ms, sys: 2 ms, total: 149 ms
Wall time: 153 ms

[lopuhin]

looks like it's possible to make it about 10x-20x faster, and have just the first run slow, by fixing how caching is performed in the dictionary module, and being smarter about stripping the tz (pop_tz_offset_from_string), which is done for every language (it's quite slow), and a locale is checked even if result is the same.