A simple assumption is made about the tags: it's assumed that they come in
chronological order, and share the same history (eg. v0.0.1, v0.0.2). That way,
to get commits made for v0.0.2, we simply look for history from v0.0.1 to
v0.0.2.

Some data from linux-2.6 and wine repositories:

linux-2.6:

Execution time 8574.73483 secs, 8539.54468 secs (99.59 %) in external commands)
Execution time 532.61431 secs, 509.50925 secs (95.66 %) in external commands)

-> time reduced to ~6%

wine:

Execution time 1538.62315 secs, 1512.65805 secs (98.31 %) in external commands)
Execution time 341.02903 secs, 331.91908 secs (97.33 %) in external commands)

-> time reduced to ~22%

And this is just one particular function that is cached.

The downside is more used diskspace, the cache for linux-2.6 for example
is 2.9 MiB at the moment.

Only names are shown for the next 5.

Two executions of 'git-rev-list' were merged and python used instead of sh
snippet. From initial testing this seems to _slow_ things down.

Get total lines of code from "git-log --shortstat" instead of "cat".

Headers are now created using html_header() function.

This name change was done just in case the statcvs/statsvn folks plan on making
a git version at some point in the future (the projects seem to share
developers).

Also fixed the script to find gnuplot files with glob.glob().