feat: X-SAMPA / Praat / SIL / Branner → IPA phonetic-alphabet modifiers

[willwade]

Summary

Adds four new SpeechMarkdown text-modifier keys — xsampa, praat, sil, branner — that translate to Unicode IPA via the ipa-translate crate and emit the same <phoneme alphabet=\"ipa\" ph=\"…\">text</phoneme> shape as the existing ipa: modifier. Sits behind a default-on phonetic-translation Cargo feature so downstream consumers can opt out of the extra dependency.

(pecan)[xsampa:\"'pi.k{n\"]
  → <phoneme alphabet=\"ipa\" ph=\"ʲpi.kæn\">pecan</phoneme>

(hello)[xsampa:\"h@\"loU\"]
  → <phoneme alphabet=\"ipa\" ph=\"həˈloʊ\">hello</phoneme>

• Google strips the phoneme tag (mirrors existing ipa: behaviour); Alexa and Azure emit it.
• Conversion is infallible per upstream — invalid source produces garbage IPA, matching how ipa: passes its value through unvalidated today.
• Integration test gates the new fixtures on the feature flag so both feature configurations stay green.
• Fixtures live in the submodule on willwade/speechmarkdown-test-files@feat/expand-test-fixtures (six new directories: xsampa-standard, xsampa-stress, praat-standard, sil-standard, branner-standard, combo-xsampa-emphasis).

Test plan

• `cargo test` (default features) — should pass including the new `phonetic_alphabet_tests` module and the six new integration fixtures
• `cargo test --no-default-features` — should pass; the six feature-gated fixtures are skipped
• `cargo clippy --all-targets` clean
• Verify downstream (`swift-tts-wrapper` etc.) still builds against the new SpeechMarkdownRust binary with the default feature on