update Roots

This update addresses a few things:

• it allows the use of Unitful values. The x values are now expected to be of type Number and not the more restrictive Real. This allows complex values to be used without special casing (as was done with Newton), though makes bisection methods throw a different error than a method not found one.
• it updates documentation to highlight the find_zero interface and not the MATLAB inspired fzero one
• it removes support for passing in a bracket to the algorithm to ensure an algorithm doesn't escape. For such algorithms, the FalsePosition method offers several variants.
• it streamlines, a bit, the setting up of a problem to pass off to find_zero.
• it modifies the tolerance used to assess f(x_n) \approx 0
• it changes the names of the tolerances to mirror that of isapprox. The older names should still work.