[codex] Fix Read Aloud setup fallback readiness

[avifenesh]

Refs #234.

What changed

This tightens the Read Aloud setup path so the settings UI reports readiness only when a speech backend can actually run:

• persist Kokoro Python/model/voices paths after the download flow
• pass CODEX_LINUX_READ_ALOUD_KOKORO_PYTHON into the app-side Kokoro runner
• fall back from the staged runner path to kokoro-stdin on PATH
• add a small Linux audio environment repair for spawned speech commands when XDG_RUNTIME_DIR is missing
• make native spd-say / espeak-ng fallback available by default inside the already opt-in feature, with CODEX_LINUX_READ_ALOUD_NATIVE_FALLBACK=0 as the opt-out
• stop forcing spd-say -t female1; use the distro default unless the user explicitly sets CODEX_LINUX_READ_ALOUD_VOICE_TYPE
• send a User-Agent for Hugging Face model downloads from both the UI helper and shell installer
• keep Read Aloud controls only in the dedicated Read Aloud settings section, and remove/clean up the older normal-settings injection path

Why

A user report in #234 showed the setting could be enabled and the model installed manually, but clicking the speech icon still produced no sound. The previous setup flow could also return ok: true after only selecting model files, even when Python, aplay, the runner, or a native fallback were not available.

The settings UI also had duplicate surfaces: the dedicated Read Aloud section plus a normal/general settings row. This keeps the feature in one clear settings location.

Validation

• node --test linux-features/read-aloud/test.js linux-features/read-aloud-mcp/test.js
• cargo test -p codex-read-aloud-linux
• cargo check -p codex-read-aloud-linux
• bash -n linux-features/read-aloud/install-kokoro-runtime.sh
• node --check linux-features/read-aloud/patch.js
• git diff --check
• node --test linux-features/*/test.js