Make GTZAN dataset sorted and use on-the-fly data in GTZAN test

[mthrok]

1. Change GTZAN to return samples in order sorted by file paths.
2. Change test to use on-the-fly data.

Relates to #764 and #791

[codecov]

Codecov Report

Merging #819 into master will increase coverage by 0.18%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #819      +/-   ##
==========================================
+ Coverage   89.87%   90.06%   +0.18%     
==========================================
  Files          35       35              
  Lines        2716     2717       +1     
==========================================
+ Hits         2441     2447       +6     
+ Misses        275      270       -5     

Impacted Files	Coverage Δ
torchaudio/datasets/gtzan.py	88.05% <100.00%> (+7.75%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3cdcd7b...62d15f1. Read the comment docs.

[vincentqb]

LGTM, but I'd say the long term solution is to align GTZAN's implementation with the other ones also, and fix all at the same time.

[mthrok]

thanks!

[mthrok]

LGTM, but I'd say the long term solution is to align GTZAN's implementation with the other ones also, and fix all at the same time.

Agreed. We need to find a generic way to perform pattern matching on paths regardless of platform.