There is at least one known issue in this version of the bridge descriptor
sanitizer:  Bridge IP addresses contained in reject lines are not
rewritten as 127.0.0.1.  There may be further issues.  People should use
the bridge descriptor sanitizer that is part of metrics-db instead.

Skip exit lists based on the file creation timestamp, not based on the
contained ExitAddress lines.  The output format is now almost the same as
the one produced by Visitor.java.

Robert Ransom points out that TAILS 0.5, and presumably other
installations of Torbutton 1.2.5, produce a user-agent string with
'chrome://global/locale/intl.properties' instead of a country/language
code like 'en-US'. Adding this user-agent string, so that we can identify
all Torbutton users.

When parsing an Apache log line, we're trying to find the last ExitAddress
line preceding the log timestamp. bisect_left gives us the insertion point
to maintain sorted order when inserting a new element. We need to make
sure this insertion point isn't index 0, meaning there's no smaller
timestamp. And we need to look at index pos - 1 to compare the log
timestamp with the exit list timestamp.

Note that exonerator.py still has the same bug!

No longer assumes Apache time zone == server time zone. Also, the header is fixed to reflect proper licensing terms.

Fixed the issue of not converting the apache log datetime to UTC. Also made a couple of improvements suggested by Damian Johnson. Next up: parallelizing get_exitlist to make the program less IO bound

Found by Kiyoto Tamura.