Measure and reduce code size of core icu_capi functionality

[sffc]

Currently the core cdylib for icu_capi is about 100 kB when generated with this command:

cargo +nightly rustc --crate-type cdylib --profile=release-opt-size --no-default-features --features looping_panic_handler,libc_alloc,buffer_provider  -p icu_capi -Zbuild-std=core,alloc -Zbuild-std-features=panic_immediate_abort --target=x86_64-apple-darwin

(on Mac, also pass RUSTFLAGS=-Clink-args=-Wl,-undefined,dynamic_lookup,-lSystem)

Looking at the cdylib, the biggest contributors to code size are

• ICU4XDataProvider_enable_locale_fallback
• icu_locid::extensions::Extensions::try_from_iter
• icu_provider_adapters..fallback..LocaleFallbackProvider
• ICU4XLocale_canonicalize
• ICU4XLocale_total_cmp
• icu_provider_blob..blob_data_provider..BlobDataProvider

Ideally we should get this core logic to be below 50k or less. There may be some low-hanging fruit.

We should add a CI job similar to the tiny jobs.

CC @robertbastian @mosuem