Use static arrays for eigen-calculations

[KnutAM]

Won't directly solve any issues, but rather move the responsibility of maintaining eigenvalue calculations to StaticArrays.jl
#173 and potentially #174/#175 (provided ForwardDiff 575) could be solved by this.
Gives less code base to maintain
Benchmarks to come soon...

[KnutAM]

Benchmark (after b4f893a)

using StableRNGs, BenchmarkTools, Random, Tensors
Base.rand(rng::AbstractRNG, ::Type{TT}) where{TT<:SymmetricTensor} = TT(ntuple((_)->rand(rng), Tensors.n_components(TT)))
rng=StableRNG(1);
for dim=1:3
        a=rand(rng,SymmetricTensor{2,dim})
        println("dim=$dim");
        @btime eigen($a);
        @btime eigvals($a);
        @btime eigvecs($a);
end

dim	eigen	eigvals	eigvecs	version
1	0.800	0.900	0.800	master
1	1.000	0.800	1.000	PR
2	6.700	5.800	6.000	master
2	6.700	1.200	5.800	PR
3	488.205	488.205	488.718	master
3	513.542	43.088	513.021	PR

So there is a 5-6 % performance hit for 3-dimensional eigenvector calculation.
But the advantage with the PR's approach is that StaticArrays' eigvals calculation could be used by adding a single line of code, improving the performance for calculating only eigenvalues

Note on 4th order
There is also a potential performance benefit of using StaticArrays for 4th order, 2-dimensional tensors (although this shouldn't be so common to do in a performance critical code?) Excluding the conversion, StaticArrays is a lot faster and without allocations.
481.538 ns (0 allocations: 0 bytes) compared to 7.450 μs (86 allocations: 7.92 KiB)
But the conversion is more tricky (not necessarily very costly), requiring e.g. a tomandel for creating static arrays.
This conversion might be useful in other cases too, but I think it makes sense to consider a separate PR...

[KristofferC]

Seems like an OK performance hit to take.

[codecov-commenter]

Codecov Report

Merging #182 (ff01fec) into master (2e9ecf2) will decrease coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #182      +/-   ##
==========================================
- Coverage   97.91%   97.84%   -0.07%     
==========================================
  Files          16       16              
  Lines        1341     1206     -135     
==========================================
- Hits         1313     1180     -133     
+ Misses         28       26       -2     

Impacted Files	Coverage Δ
src/Tensors.jl	81.81% <ø> (ø)
src/math_ops.jl	100.00% <ø> (ø)
src/eigen.jl	100.00% <100.00%> (+1.40%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2e9ecf2...ff01fec. Read the comment docs.

[KnutAM]

I can remove to_smatrix(::Tensor) (not used) and add test for eigvecs before it is ready for merge!
I'll look into that tomorrow!

[KnutAM]

Ready from my point of view now!
Let me know if I should change something more!