Add validation to log-regression benchmark

[lovisek]

Validation currently supports only binary classification which returns a single intercept value. If multinomial logistic regression is needed, the validation needs to be updated to support multiple intercept values.

The validation checks logistic regression model's coefficients, intercept, and the number of classes the model can classify. For the coefficients, it compares the sum and sum of squares of all coefficient values with the expected values, and then validates the coefficient count. Regarding the intercept, its value is compared with the expected value, and then the validation checks that the model contains only one intercept.

The validation does not change the code that is being measured and should not really impact benchmark performance. As a quick check, I calculated an average duration from the last 5 repetitions of a single run of the benchmark before and after validation:

• without validation: 920,128ms
  log-regression.no-validation.result.txt
• with validation: 929,781ms
  log-regression.with-validation.result.txt

The variant with validation appears approx. 1% slower, but this is just from a single run. Given that the benchmark code did not change, I don't think this indicates an actual change, but if necessary, I can do the comparison for multiple runs.

[farquet]

LGTM!
Thanks for the contribution.