[Backend Tester] Add tensor error statistic reporting

[GregoryComer]

Report various error statistics for the test outputs, including SQNR, mean absolute error (MAE), and L2 norm. These are saved in the detail report per test case.

As an example, here is the output from Core ML running MobileNet V2 (roughly formatted from csv -> sheets -> markdown):

Output 0 Error Max	Output 0 Error MAE	Output 0 Error MSD	Output 0 Error L2	Output 0 SQNR
0.0005887411535		0.0001199183663		2.32E-06		0.004750485188		41.28595734

TODO: Round to a reasonable number of significant figures. (Or leave in full precision in the report?)

[GregoryComer]

Stack from ghstack (oldest at bottom):

• [Backend Tester] Report delegation statistics #12846
• [Backend Tester] Report quantization and lowering times #12838
• -> [Backend Tester] Add tensor error statistic reporting #12809
• [Backend Tester] Add CSV report generation #12741
• [Backend Tester] [WIP] Add Qualcomm tester and register flow #12739

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/12809

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit ef7af5c with merge base 4fd2079 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[GregoryComer]

I'm not completely happy with the callback approach for exposing this, but I don't really have a better idea, since the tester relies on a builder-style pattern where it returns self to allow chaining. I'm open to suggestions.

[digantdesai]

just update the tester method?

[GregoryComer]

Can you clarify what you're thinking? Are you meaning update the tester run_method_and_compare outputs to directly return the error stats and then update all of the callers to not use it in a chained fashion? Or something else?

[digantdesai]

Modify the existing method, keeping the outside behavior but also add def get_comparison_stats(self) method on that stage or something?
Or if you want to pass a callback for flexibility that's also fine.

[GregoryComer]

I'll add some tests for the error statistic calculation to this PR shortly.