Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@crates/stdlib/src/unicodedata.rs`:
- Around line 44-55: The code mixes Unicode data sources:
Age::of()/UNICODE_VERSION (unic_ucd_age, Unicode 10) vs. ICU-based APIs
(icu_properties/icu_normalizer, Unicode 17), so update the module to use one
consistent source—preferably the ICU data already used by category(),
bidirectional(), east_asian_width(), normalize(), is_normalized(), and
mirrored(): remove references to unic_ucd_age::Age and UNICODE_VERSION, derive
unidata_version from the ICU provider/metadata (the same runtime data backing
icu_properties and icu_normalizer), and change check_age() to use ICU's age
information (or remove age-based filtering) so that unidata_version accurately
reflects the data used by those functions. Ensure all places that previously
consulted Age::of() now query the ICU data provider or use the ICU-provided
UnicodeVersion so behavior and reported unidata_version remain in sync.

---

Nitpick comments:
In `@crates/stdlib/src/unicodedata.rs`:
- Around line 194-217: The normalize function duplicates the
map_utf8(...).collect() pipeline for each NormalizeForm branch; instead, choose
the appropriate normalizer constructor based on the form (use the variants Nfc,
Nfkc, Nfd, Nfkd and the constructors ComposingNormalizerBorrowed::new_nfc(),
ComposingNormalizerBorrowed::new_nfkc(),
DecomposingNormalizerBorrowed::new_nfd(),
DecomposingNormalizerBorrowed::new_nfkd()) and store it in a single variable (or
closure) and then call text.map_utf8(|s|
normalizer.normalize_iter(s.chars())).collect() once; apply the same refactor to
the later duplicate block as well so only the normalizer choice varies and the
shared map_utf8(...).collect() path is centralized.

Verify each finding against the current code and only fix it if needed.

Outside diff comments:
In `@crates/stdlib/src/unicodedata.rs`:
- Around line 179-190: The comment above the east_asian_width method is outdated
(mentions "uses 9.0.0 database") but the implementation now relies on ICU
property data; update the note above fn east_asian_width to accurately state
that EastAsianWidth values are derived from ICU property data (or similar
current source) instead of referencing the old 9.0.0 database, ensuring the
comment matches the implementation without changing code logic.
- Around line 164-176: The bidirectional method currently returns ICU
abbreviations via BidiClass::short_name(), which yields "BN" for U+FFFE; update
bidirectional (in the bidirectional function where you call self.extract_char
and BidiClass::for_char/short_name) to explicitly return an empty string for
U+FFFE (or other noncharacter codepoints that CPython treats as no bidi) instead
of the ICU short name — i.e., after extracting the char, check its codepoint
(c.to_char().map or similar) and if it equals 0xFFFE (or otherwise identified as
a noncharacter per CPython behavior) return "" before calling
BidiClass::short_name().

---

Nitpick comments:
In `@crates/stdlib/src/unicodedata.rs`:
- Around line 194-247: The normalize()/is_normalized() functions duplicate the
same NormalizeForm -> normalizer construction; extract that dispatch into a
private helper (e.g., fn get_normalizer(form: super::NormalizeForm) -> impl
Fn(&str) -> impl Iterator<Item=char> or return an enum/trait object representing
the chosen normalizer) that maps Nfc/Nfkc to
ComposingNormalizerBorrowed::new_nfc()/new_nfkc and Nfd/Nfkd to
DecomposingNormalizerBorrowed::new_nfd()/new_nfkd, then call that helper from
both normalize() and is_normalized() to run text.map_utf8(|s|
normalizer.normalize_iter(s.chars())).collect() and compare results, removing
the duplicated match blocks.