Handle browsers with no navigator.language better (#521)

[zackw]

The code for setting config["lang"], in isso/app/config.js,
assumes that browsers will always provide a value for
navigator.language and/or navigator.userLanguage. Per bug #521,
this is not a safe assumption.

While I was attempting to fix this, I noticed that regional variants
of a language (zh-TW, pt-BR) were being handled in an ad-hoc,
unreliable manner. I also noticed a new user-agent language property
navigator.languages which more closely matches the semantics of
Accept-Language—it would be good to support that.

This patch addresses all the above problems, as follows:

1. Add a new configuration property data-isso-default-lang that
   specifies the language to use (instead of English) when the browser
   doesn’t have a preference.

2. Document that we expect the value of data-isso-lang and
   data-isso-default-lang to be a BCP 47 language tag, because
   this is what navigator.language etc contain. (The practical
   upshot is that tags like zh-TW are officially allowed now.)

3. In config.js, compile an array of candidate language tags, in
   descending order of preference, from all available sources:
   data-isso-lang, navigator.languages, navigator.language,
   navigator.userLanguage, data-isso-default-lang, and finally
   "en" as a backstop. Handle any or all of the above being
   null/undefined/empty. This array goes into config["langs"].
   config["lang"] is removed.

4. In i18n.js, select the first entry in config["langs"] for which
   we have both pluralforms and a translation, and make that the value
   of i18n.lang. An English backstop is supplied here too for extra
   defensiveness. Also, if we don’t have a translation for say
   zh-HK, try zh before moving on to the next entry in the array.

5. New function utils.normalize_bcp47 ensures that we process
   language tags, whereever they came from, case-insensitively and
   treating _ as equivalent to -.

6. Move utils.ago to i18n.ago to avoid a circular dependency
   between utils and i18n.

[ix5]

Good job on the PR!

I've tested with navigator.lang{ueages} set to en, en-US, de, de-de and pt-BR so far.
Apart from the nit in the linline comment I posted, it works great.

Could you rebase on latest master please? The pt_PT translation was added in the meantime.
I've pushed a rebased version with the fix for lang is undefined to ix5/isso:navigator-lang-674 for you to pick.

[zackw]

@ix5 Rebased and made both of the requested changes.

[ix5]

Great, thanks! @jelmer @blatinier please review when you find the time.

[ix5]

@zackw I've rebased your branch on top of current master, please see https://github.com/ix5/isso/tree/pull-674-rebased
Would you mind adding an entry to the CHANGES.rst file as well?

After that, I'll merge this PR.

[zackw]

@ix5 Done.

[stefangehn]

LGTM, I think proper handling of the two-part language identifiers is also worth merging this.

[ix5]

Thank you for reviewing, @stefangehn!
And thank you for your patience @zackw!