Performance improvements with text processing

[timholy]

I don't have time to make this a PR right now, so this is just a placeholder so it doesn't get lost: see https://groups.google.com/d/msg/julia-users/2uaRs3JIdfw/hMLdj6wxCwAJ. The most important change was making the string encoding a type parameter of what is effectively EachLine, and then using that in downstream code to prevent type instability.

Without this, on the "Hungarian Wikipedia" test data set (see https://github.com/juditacs/wordcount), some lines come back as UTF8String and some as ASCIIString. Might as well keep it consistent.

[JeffBezanson]

ref #1792

[nalimilan]

Looks like this is another issue which will get fixed by merging ASCIIString and UTF8String into a single String type (#14383), which should get rid of type instability in EachLine.

@timholy What other lessons to you draw from that thread regarding thread processing? I must say it's not immediately obvious to me.

[timholy]

Thanks for the reminder. I have a grant deadline of Wednesday, after which I'll try to remember to come back to this.

[nalimilan]

No worries. I was just trying to link related issues together.

[timholy]

Fixed by the strings changes.