Improve support for linear algebra

[OlivierHnt]

Some linear algebra functions are accessible from Base, e.g. * and inv.
This PR implements methods for these functions as part of the LinearAlgebra extension to avoid getting error messages and the need to load the more specialised package IntervalLinearAlgebra.

Specifically,

• adds a simple matrix inversion algorithm based on the contraction mapping theorem (x-ref: [enhancement]: inv and det (needed by IntervalRootFinding) IntervalLinearAlgebra.jl#134)
• implements Rump's algorithm (probably the most naive one, which does not rely on OpenBlas to avoid multithreading issues, x-ref: [enhancement]: Ship a correctly rounded threaded OpenBLAS as an Artifact IntervalLinearAlgebra.jl#131)

It also defines some methods for the operator norm (opnorm) since the default algorithm used the syntax sum::Tsum = 0 which leads to an unnecessary "NG" flag.

I took the tests for matrix multiplication from IntervalLinearAlgebra.

cc @lucaferranti @orkolorko