Fix torch.randn to use proper native function

[davoclavo]

Fix subtle bug with torch.randn, as it was mistakenly using rand under the hood.

I tried implementing a variance check as well, but I couldn't find the var function under javacpp. If anyone is able to help me I'd appreciate it.

As a side note, I will keep implementing more tensor ops soon, and improving test coverage to try to avoid these kind of subtle errors; a handful of them happened to me while implementing pointwise ops.

[davoclavo]

Update: I found the variance function, it is under nativeTorch.`var` - I will implement that function + the unit test check, so please don't merge yet

[davoclavo]

Done! I just added Tensor.variance + relevant check in the randn unit test

I hesitated sticking to the pytorch nomenclature due to the forced backticks in Scala (Tensor.`var` ) because it feels unnatural, and would require translating those functions names anyways (var -> `var`). However, please let me know I should implement it with pytorch's name (and backticks), or if I should add it under both names, happy to do either one.

[sbrunk]

Looks good to me, thanks!

I hesitated sticking to the pytorch nomenclature due to the forced backticks in Scala (Tensor.`var` ) because it feels unnatural, and would require translating those functions names anyways (var -> `var`). However, please let me know I should implement it with pytorch's name (and backticks), or if I should add it under both names, happy to do either one.

Also in favor of just using variance here. I think it helps with discoverability. If you type var in the IDE or in Scaladoc search and it should suggest variance

As a side note, I will keep implementing more tensor ops soon, and improving test coverage to try to avoid these kind of subtle errors; a handful of them happened to me while implementing pointwise ops.

Awesome! FYI just to avoid double work, I've started implementing reduction ops in #28