Default languages strategy for Ubuntu desktop CD

Even if ubiquity is doing an awesome job at installing on demand languages on ubuntu, having more default languages as we used to on the destkop image enables us to have the live session localized. Also, offline installation will bring that support without adding supplementary download time.

Some loco team used to have respin of the default localized ubuntu image. Yours truly used to have handle this quite regularly for the french loco team. I know that the italian and finish locos used to do it as well. (not sure about the current status for them) However, UEFI support made it complicated and so those efforts have been dropped some cycles ago as far as I know.

I would thus propose that we enhance and complete our default language strategy for xenial. The LTS is an excellent time to make progress on this.

The proposal is to:
1. Install full language support for those shipped on xenial image. It means that opening "language selector" won't request any additional package to install[1]. If you are proceeding an online installation, additional packages won't be downloaded to complete your language installation. If you have done an offline one, you won't have the infamous after first boot "Language support is not complete" dialog. Note that for now, we have no complete language support on the live! For instance, in English, we have the following missing packages that language-support will require to install (or that ubiquity will download it for you if you are connected to the Internet):
hyphen-en-us, mythes-en-us, mythes-en-au, hunspell-en-ca, myspell-en-au, myspell-en-gb, myspell-en-za, libreoffice-help-en-gb, libreoffice-l10n-en-za, firefox-locale-en, thunderbird-locale-en, thunderbird-locale-en-gb, thunderbird-locale-en-us.

2. Based on popcon, number of native speaker and total number of speaker per language, it seems that the following language selection makes sense for our user base (more info on the language selection below):
en, es, zh (simplified), pt, de, fr, it, ru

3. We remove the difference between archs for now on the ubuntu desktop live seed (no language installed only on i386/powerpc like German and Portuguese, where they aren't installed on amd64), no more partial libreoffice help support only installed on i386 and amd64, for consistency.

Additional estimated size on the live to achieve this is 95Mb adding the full support for those 8 languages. This is of course a non negligeable image size increase. However, getting full language support for at least 8 languages (and no additional download on install for those) seems like a huge increase of quality to me. I'm sure as well that with the python2 drop support and other cleanup efforts, we would be able to regain a great part of that additional iso size (remember that our image was still at 1Gb not so long ago where we are at 1.2Gb now).

Finally, keep in mind that those locales are only part of the live image. The chosen language will be present on the installed system, but no other ones, even if shipped on image. So no extra space will be required on the installed target.

[1] apart from one font for simplified chinese which is weighting 8Mb.

-----
Full rationale on what to install, why and languages selection:

Back in the day we were still producing Ubuntu desktop ISO being able to burn on the CD, we had to fight a lot every cycle to keep the size reasonable, while still having a reasonable offer in term of default languages on the CD. However, at some point, this wasn't possible anymore and we dropped many languages as our default installation.

We lifted this image size restriction some cycles ago (and decided that the ISO will be used on DVDs or USB sticks) but didn't reintroduce languages that we dropped by then.

More and more locos (in particular Italian and French) don't do their own ISO respin anymore due to issues with UEFI.

In addition, some CD image time decision are still on the live (like german and portuguese are installed on all archs but amd64… which is the most widespread archs). We should use this time to cleanswap this.

Finally, when we install support for one language, we should install it fully. It means that opening "language selector" should tell you that the language is fully installed for the default application selection/dictionaries and that you don't rely on an Internet connexion to complete your locale installation. Even the default english selection
isn't complete and miss a bunch of packages. For instance, on xenial:

For this, we are going to use /usr/share/language-selector/data/pkg_depends (used by language-selector) and shipping them in live seed task for every dependent package that we shipped by default.

I propose we try and assess around the following criteria to decide what to install by default:
- Number of native speaker of one language: https://en.wikipedia.org/wiki/List_of_languages_by_number_of_native_speakers
- Popularity contest to assess what languages are using our community the most: http://popcon.ubuntu.com/by_vote. This as well will enable us to see communities (I think German people as an example) which are traditionally using their systems in English, contrary to others (italian, french), which are switching locale to their native languages.
- Previous ubuntu versions we shipped locale for (I propose we look at 10.04 LTS)
- Additional size on the ISO. We will approximate packagesize ~= size taken once extracted files are compressed in the squashfs.

I'm linking a seed branch fixing those issues. Note that ubiquity will need some checking to ensure that all desired packages are indeed installed even offline.