Small improvements to python scripts

[rejunity]

Changes are aimed to reduce additional steps when benchmarking on a fresh setup and makes it easier to update README.md file.

1. Automatised cuDDN version conversion from int to a friendly string. Will automatically convert 5105 to '5.1.05'. Implemented as cudnn_name() function instead of cudnn_map dict.
2. Formatted printed out tables so that they are more readable in text format. That does not affect .md formatting since .md parser strips whitespace.

BEFORE:

|Maxwell Titan X|4.0.07|150.90|268.64|419.54|
|Pascal Titan X|None|238.04|371.40|609.43|
|GTX 1080 Ti|None|225.36|368.42|593.79|
|GTX 1080|None|299.05|461.67|760.72|
|Maxwell Titan X|None|382.39|583.83|966.22|
|CPU: Dual Xeon E5-2630 v3|None|6572.17|10300.61|16872.78|

NOW:

|Maxwell Titan X          |4.0.07 | 150.90| 268.64| 419.54|
|GTX 1080 Ti              |None   | 225.36| 368.42| 593.79|
|Pascal Titan X           |None   | 238.04| 371.40| 609.43|
|GTX 1080                 |None   | 299.05| 461.67| 760.72|
|Maxwell Titan X          |None   | 382.39| 583.83| 966.22|
|CPU: Dual Xeon E5-2630 v3|None   |6572.17|10300.61|16872.78|

3. Made order of printed out models to match order in README.md. Order is defined by model_names_sorted array in analyze_cnn_benchmark_results.py file
4. Automatically create output directory, if missing.
5. Changed DEFAULT_MODELS to match file structure in models.zip