Provide a no-op `convert`

[timholy]

This should help performance, but it also resolved a subtle bug that crops up if you perform autodiff with respect to AngleAxis coordinates at the origin: sqrt(dual(0.0)) gives you NaNs. Those cropped up because AngleAxis doesn't remember if you called it with normalize=false.

[codecov[bot]]

Codecov Report

Merging #200 (ef874f4) into master (15d1c32) will decrease coverage by 0.28%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #200      +/-   ##
==========================================
- Coverage   86.12%   85.84%   -0.29%     
==========================================
  Files          14       14              
  Lines        1341     1342       +1     
==========================================
- Hits         1155     1152       -3     
- Misses        186      190       +4     

Impacted Files	Coverage Δ
src/core_types.jl	89.62% <100.00%> (-0.67%)	:arrow_down:
src/euler_types.jl	91.68% <0.00%> (-0.74%)	:arrow_down:

---

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 15d1c32...ef874f4. Read the comment docs.

[hyrodium]

autodiff with respect to AngleAxis coordinates at the origin: sqrt(dual(0.0)) gives you NaNs.

Just a comment, is there any case to calculate differentiation on the origin? I think the vector of AngleAxis is assumed to be of unit length.