Optimize data size for likely subtags

[sffc]

As discussed in #1462, I would like to consider making likely subtags a central feature of locale resolution when performing vertical fallback in data loading.

However, the current 40 KB size for the likely subtags data is a rather heavy dependency to take on. We should investigate:

1. Enabling the data to be sliced by language/region
2. Reducing the absolute size of the data by, e.g., removing duplicated subtags between input and output, and by removing "und" in places where it does not matter
3. Consider a mode that does not add in the script subtag if the script is the default for the language (e.g., considering "zh-CN" to be maximized, rather than "zh-Hans-CN"), which may enable us to reduce the size of some mappings

CC @zbraniecki

[mihnita]

if the script is the default for the language

Yes, but then you need that info somewhere :-)
And might not save that much.

Here are all entries that map from / to zh:

<likelySubtag from="zh" to="zh_Hans_CN"/>
<likelySubtag from="und_030" to="zh_Hans_CN"/>
<likelySubtag from="und_142" to="zh_Hans_CN"/>
<likelySubtag from="und_CN" to="zh_Hans_CN"/>
<likelySubtag from="und_Hans" to="zh_Hans_CN"/>

<likelySubtag from="zh_Bopo" to="zh_Bopo_TW"/>
<likelySubtag from="zh_Hanb" to="zh_Hanb_TW"/>
<likelySubtag from="und_Bopo" to="zh_Bopo_TW"/>
<likelySubtag from="und_Hanb" to="zh_Hanb_TW"/>
<likelySubtag from="und_Hani" to="zh_Hani_CN"/>

<likelySubtag from="zh_AU" to="zh_Hant_AU"/>
<likelySubtag from="zh_BN" to="zh_Hant_BN"/>
<likelySubtag from="zh_GB" to="zh_Hant_GB"/>
<likelySubtag from="zh_GF" to="zh_Hant_GF"/>
<likelySubtag from="zh_HK" to="zh_Hant_HK"/>
<likelySubtag from="zh_ID" to="zh_Hant_ID"/>
<likelySubtag from="zh_MO" to="zh_Hant_MO"/>
<likelySubtag from="zh_PA" to="zh_Hant_PA"/>
<likelySubtag from="zh_PF" to="zh_Hant_PF"/>
<likelySubtag from="zh_PH" to="zh_Hant_PH"/>
<likelySubtag from="zh_SR" to="zh_Hant_SR"/>
<likelySubtag from="zh_TH" to="zh_Hant_TH"/>
<likelySubtag from="zh_TW" to="zh_Hant_TW"/>
<likelySubtag from="zh_US" to="zh_Hant_US"/>
<likelySubtag from="zh_VN" to="zh_Hant_VN"/>
<likelySubtag from="zh_Hant" to="zh_Hant_TW"/>
<likelySubtag from="und_HK" to="zh_Hant_HK"/>
<likelySubtag from="und_MO" to="zh_Hant_MO"/>
<likelySubtag from="und_TW" to="zh_Hant_TW"/>
<likelySubtag from="und_Hant" to="zh_Hant_TW"/>

So we would save more space if we drop Hant, not Hans.

[robertbastian]

I'll do (2) as part of #1034

[sffc]

The low-hanging fruit is done. The rest of this issue will require some deeper design.

[sffc]

I don't think there is much left to do here. Our likely subtags data is now very small, especially with the core and extended slicing.