Add arguments specifying subsets of eigenvalues in eigfact(Symmetric/Hermitian)

[andreasnoack]

No description provided.

[ViralBShah]

Is it possible to directly show the output from travis in the PR? Cc: @loladiro

[Keno]

I'm not sure what you're asking.

[JeffBezanson]

From the docs, it was not clear to me how to specify that I'm passing il, iu as opposed to vl, vu. This seems to be based on whether the arguments are integers, which strikes me as extremely bad. Integers are also perfectly good numbers.

[jiahao]

This syntax also exists in the stein! wrappers for SymTridiagonal eigenproblems. In retrospect, perhaps a more Julian syntax would be to specify the interval [vl, vu) as two numeric arguments and the range il:iu as a UnitRange. (The latter is semantically an ordinal range, specifying the ilth through iuth sorted eigenvalues, while the former is a semiopen interval.)

[JeffBezanson]

That would be an improvement. The way it is now is really confusing, and will break in odd ways if the type of one of the numbers is not what you expect.

[andreasnoack]

Good idea. I'll change the Int method to a method for UnitRange and the Float64 method to a Number method.

[andreasnoack]

Done in #6678