Implement StandardScaler; add associated tests

[cryptodeal]

Implemented BaseScaler abstract class + StandardScaler, which extends the base class. Also added simple unit tests for StandardScaler.

Fixed a few type errors in shumai/tensor/tensor.ts while working on this.

[cryptodeal]

Also, noticed that Bun's GC is eager to the point that when running consecutive bun wiptest, I would see an occasional failure of the test on memory as Bun was garbage collecting tensors eagerly enough that the resulting mem usage was less than at start as Bun had GC'd some tensors before the call to dispose.

Accordingly, I've updated the test to check that the memory is less than or equal to memory usage at the start of the test.

[asilvas]

LGTM just a couple suggestions

[cryptodeal]

Biggest issue is the weirdness w test CI runs failing due to mismatched dtype w/ logged warnings RE usage of Float16Array from this test (when the tests don't explicitly use Float16Array). I'm going to mark as a draft pending further investigation into this. (Unable to replicate that error specifically locally, but there's further bugs I'm finding where a test will pass when run as Float32Array and fail when run as Float64Array, so definitely needs to be debugged before ready to merge.

Tweaked implementation to debug a few errors I was catching locally.

[cryptodeal]

Going to implement remaining class methods so that it's at feature parity with the sklearn.preprocessing.StandardScaler implementation, then will flag as ready for review.

[cryptodeal]

At this point, I think there's either something I'm missing in the implementation causing failure on Linux runs due to transformed.dtype not matching the original dtype of the original Tensor.

Welcome feedback with regard to the implementation; I have largely directly ported this Golang sklearn port as I'm more familiar with Go than python.

[cryptodeal]

Currently, it's failing on GPU when comparing dtype of scaled against dtype of the pre-transformed inputs. The actual values are scaled as expected, but dtype doesn't match.