docs: refresh homophones README — (name, prob, similarity) + per-lang…

…uage g2p2 (#25)

Document the 3-tuple return, correct the stale 'single English G2P' notes
(now per-language g2p_ipa), and update the balanced-sweep description.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

feat(homophones): re-sweep balanced weights on per-language g2p phone…

…tics (#23)

- sweep_balanced.py: IPA term now uses g2p2.similarity(weighted) on the
  g2p_ipa column (per-language) instead of char levenshtein on anglocentric IPA.
- faker2.naming.realnames: load the g2p_ipa column (parity with Rust).
- data/balanced_params.json: re-swept over 119 countries — with reliable
  per-language phonetics the acceptance threshold settles higher (min 0.6).

Balanced stays clean: Sophie->Sophia/Sofia, Mohammed->Muhammad, Giovanni->Gianni.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

feat(homophones): per-language g2p2 phonetics for the whole name DB (#22

)

Phonemize all 1.43M names in their own language with g2p2 (99 countries -> 72
languages, en fallback for romanized/unsupported) and store as a `g2p_ipa`
column (+ `g2p_lang`). Homophones now compute similarity on these real
per-language phonemes instead of the anglocentric IPA:
  Mohammed/EG -> Mohammad/Muhammad, Giovanni/IT -> Giovanna/Gianni.

Rust realnames loads g2p_ipa (raw) and scores via g2p2's weighted alignment.
Unsupported languages tracked in g2p2#5.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

feat(rust): text formatters + pytimezone — coverage 88% -> 93% (100% …

…of string formatters) (#21)

- misc: json, csv, tsv, psv, dsv, fixed_width, xml (well-formed text output).
- python: pytimezone (IANA tz name).
- date_time: date_between_dates / date_time_between_dates aliases.

213/230 resolve via Faker::gen; the remaining 17 return raw bytes, Python
collections, or need an argument — every single-string formatter is now ported.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

feat(rust): text formatters + pytimezone — coverage 88% -> 93% (100% …

…of string formatters) (#21)

- misc: json, csv, tsv, psv, dsv, fixed_width, xml (well-formed text output).
- python: pytimezone (IANA tz name).
- date_time: date_between_dates / date_time_between_dates aliases.

213/230 resolve via Faker::gen; the remaining 17 return raw bytes, Python
collections, or need an argument — every single-string formatter is now ported.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

feat(homophones): use g2p2 articulatory phonetic similarity (#19)

Integrate jqueguiner/g2p2 (zero-dep G2P + phonetic similarity) into the
homophone scorer:
- balanced method: IPA term now uses g2p::similarity(Method::Weighted) —
  articulatory feature-weighted phoneme alignment instead of char levenshtein.
- ipa method: selects candidates by g2p similarity threshold.

Dominique/FR -> Dominik/Dominick/Domenico; Sophie -> Sophia/Sofia (no Xavier).
g2p pulled in under the real-names feature (git dep). Tests green.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

fix(pypi): correct project URLs (jqueguiner/faker2, not joke2k) (#18)

Add Repository/Issues/Changelog/Releases so the PyPI page points at this
project instead of the upstream joke2k/faker metadata.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

ci(release): PyPI publish via API token (secrets.PYPI_API_TOKEN) (#17)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

ci(release): version 40.29.0.post<run> to supersede legacy 40.28.1 an…

…d never collide (#16)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

ci(release): version 40.29.0.post<run> to supersede legacy 40.28.1 an…

…d never collide (#16)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>