Light studied the problem, but didn't solve it in his 1917 [dissertation][1].
A [1920 note][2] of Light was followed by a [1921 note][3] of Franklin showing the  existence of infinitely many classes of examples not covered by Light, and pointing out that Puiseux had shown this in 1844. However, the question of determining all of them isn't addressed there, and I don't know what else is known.

Regarding your question about formalizing the notion of curves as eigenvectors, couldn't you just take the real vector space whose basis consists of representatives from each similarity class of curves?  Scalar multiplication of a basis element by a real number corresponds to dilating or shrinking, and reflecting if negative.  The evolute operator is formally extended to this vector space linearly, and the curves similar to their evolutes will be precisely (scalar multiples of) the basis vectors that are eigenvectors.

  [1]: http://books.google.com/books?id=EtTNAAAAMAAJ&pg=PA18#v=onepage&q&f=false
  [2]: http://www.jstor.org/stable/2972254
  [3]: http://www.jstor.org/stable/2972285