Add Linear Algebra routines in ChainerX

I’d like to work on adding linear algebra routines to ChainerX during GSoC. Most of the numpy.linalg functions listed in ChainerX routines spreadsheet have low priority, so their implementation probably can wait for the summer.

Many linalg operations have optimized implementations in various open source alternatives, for example in LAPACK (there is a C++ interface). Is it expected to implement forward operations from scratch, or link to existing implementations?

For autodiff, derivatives for some routines are described here

Ref. #6423

[hvy]

#7474 has now been merged 😃

[IvanYashchuk]

Here is the table to track the progress

Routine	Forward CUDA	Forward Native	Backward	Merged?
linalg.cholesky	☑	☑	☑	yes
linalg.qr	☑	☑	☑	yes
linalg.svd	☑	☑	☑	yes
linalg.eigh	☑	☑	☑	yes
linalg.solve	☑	☑	☑	yes
linalg.inv	☑	☑	☑	yes
linalg.pinv	☑	☑	☑	yes