Improved Efficiency of the DiskUsage Metric

[vlomonaco]

Hello,

This PR speeds up the DiskUsage metric.

It turns out that if you have a lot of dirs/files in the current directory (the default target for the size inspection) especially the eval mode with disk_usage_metrics(stream=True) becomes super slow (the metric is computed every mini-batch).

In my case it takes about 4 seconds for every inspection. I use the "du" command (it improves the timing of about 4x) if you are on Linux adding a warning in case the metric compute time exceeds 0.5 seconds.

Let me know if you think I should change anything.

[coveralls]

Pull Request Test Coverage Report for Build 9268694310

Details

• 18 of 30 (60.0%) changed or added relevant lines in 2 files are covered.
• 2 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-0.004%) to 51.799%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
tests/evaluation/test_disk_usage.py	6	9	66.67%
avalanche/evaluation/metrics/disk_usage.py	12	21	57.14%

Files with Coverage Reduction	New Missed Lines	%
avalanche/benchmarks/datasets/cub200/cub200.py	1	27.55%
avalanche/benchmarks/classic/ex_model.py	1	31.13%

Totals
Change from base Build 9204029670:	-0.004%
Covered Lines:	15091
Relevant Lines:	29134

---

💛 - Coveralls