[release-0.5] revert #17389 for 0.5.0-rc1

[tkelman]

caused a big performance regression, should have ran nanosoldier before merging

[tkelman]

@nanosoldier runbenchmarks(ALL, vs = "@0030eec2f332f353e6890ca289ac2aca55532dde")

[pabloferz]

I think I found a solution for the rest of the problems, but this is a good idea anyway.

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels

[tkelman]

So since rc0, ["tuple","index",("sumelt","NTuple",30,Float64)] got almost 2x worse, a bunch of others got maybe 2x better, and a handful got 15-20% worse (but may be noise).

[KristofferC]

The native code for sumelt with tuple is identical in rc0 and master.