Canonicalize tz IDs when finding exemplar location

Canonicalize tz IDs when finding exemplar location

Canonicalize time zone IDs when looking up exemplar location.

ICU's canonical IDs are not guaranteed the same as IANA's.
Various data files (e.g. timezones.xml, tzdata,
time_zones_by_country.xml) are using IANA IDs. ICU's IDs
are often the same, but because they are guaranteed not to
change and IANAs are not there is some drift.

This fixes an issue with the settings picker for ICU IDs:

America/Buenos_Aires
Asia/Rangoon
Asia/Katmandu
Asia/Calcutta

Which are IANA's:

America/Argentina/Buenos_Aires
Asia/Yangon
Asia/Kathmandu
Asia/Kolkata

introduced by commit b4d003da66.

It could also influence some display names for
automatically-selected time zones outside of this list.

The issues was particularly noticeable in locales that
use non-ASCII scripts; ICU performs a fallback if the ID
isn't canonical and would transform and return the last
part of the ID (e.g. Buenos Aires, Yangon, Kathmandu,
Kolkata).

Bug: 36469833
Test: Manual testing in Japanese
Change-Id: I2febdbee58e474d251d0a33f2c8335664a74581f