Tweak spherical norm coefficients

[jmert]

The first commit simply eliminates a few unwanted instructions.

• Using @fastmath(sqrt) gets rid of the domain checking that the argument is positive, which eliminates one branch from the fast path.
• Multiplying by an inverse is changed to a division, which saves one floating point operation.

The second commit is more subtle. The expression for the μ_ℓ coefficient in the spherical normalization is modified to gain a few ulps of accuracy across what is probably the entire practical domain of degrees (ℓ). (N.B. the muladd subexpression does not need to be an FMA to maintain accuracy, but FMA is allowed if it is more efficient.)

Before:

julia> f1(l) = Legendre.Plm_μ(LegendreSphereNorm(), Float64, l);

julia> f2(l) = Float64(Legendre.Plm_μ(LegendreSphereNorm(), BigFloat, l));

julia> Δ(l) = f1(l) - f2(l);

julia> findfirst(!=(0), Δ.(1:500_000))
11

With this PR:

julia> findfirst(!=(0), Δ.(1:500_000))
31722

[codecov[bot]]

Codecov Report

Merging #16 into master will increase coverage by 0.01%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #16      +/-   ##
==========================================
+ Coverage   95.51%   95.52%   +0.01%     
==========================================
  Files           9        9              
  Lines         245      246       +1     
==========================================
+ Hits          234      235       +1     
  Misses         11       11              

Impacted Files	Coverage Δ
src/norm_sphere.jl	100.00% <100.00%> (ø)

---

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 a0eadb8...2210077. Read the comment docs.