Add LinearMatrix.from_mapping, and replace BladeMap with it

[eric-wieser]

For the example application in spacetime algebra, bm(x) is at least 10x faster.

In theory we could now also write split without using bm at all:

split = LinearMatrix.from_mapping([
    # even subalgebra of D
    (D.scalar, P.scalar),
    (d01, p1),
    (d02, p2),
    (d03, p3),
    (d12, p12),
    (d23, p23),
    (d13, p13),
    (d0123, p123),

    # split(odd) = split(odd * d0)
    (d0, P.scalar),
    (d1, -p1),
    (d2, -p2),
    (d3, -p3),
    (d012, p12),
    (d013, p13),
    (d023, p23),
    (d123, -p123),
])

but I don't have a good enough understanding of what split.adjoint(p1) should be to feel comfortable doing this.

[arsenovic]

so, pedagogically i dont think it adds anything to STA. however, i think its good because it teachs the transformation tools at the same time, and these are more generally useful

[arsenovic]

i like this change.