instead of summing LOC changed per file for author, average % LOC changed per file

[imsky]

this prevents long files from dominating a pull request's blame metrics, e.g.

File	Alice	Bob
1000_loc.txt	400	505
100_loc.txt	100	0
10_loc.txt	5	0
Total LOC	505	505
Avg % LOC	63%	17%

here Bob equal to Alice on LOC changed, even though Alice has much higher average ownership