Return linbasex arrays as distr object

[DanHickstein]

As discussed in #270 (comment), we should consider changing the way that linbasex returns its arrays. It should probably be more like rbasex, which uses the distr object. This way, only two variables are returned, instead of two or three.
(Note, I haven't done much research about this, I'm just opening an issue since @stggh seemed to think that this was a good idea.)

[MikhailRyazanov]

Two comments:

1. Currently my Distributions.Results class holds cosⁿ θ terms inside and converts them to other representations (such as harmonics or anisotropy parameters β_n) upon request. This works fine for low orders, but cosⁿ θ are actually very non-orthogonal (see Hilbert matrix), and thus their conversions to other representations are not well-conditioned, potentially leading to huge numerical errors when sufficiently high orders (n ≳ 10) are involved. linbasex apparently uses Legendre polynomials throughout, which are orthogonal and thus probably more numerically stable (although I don't know whether this is true in practice). In principle, Distributions.Results can be modified to hold any representation and convert it to any other when needed. I was thinking about this, but have not implemented it yet.

2. linbasex also returns projections. I don't know whether they are useful for anything, but if yes, it will need to return 3 values inv_IM, distr, projections instead of 2 values inv_IM, (radial, Beta, projections).