fix(tts): normalize markdown before speech synthesis

[rishiskhare]

Summary

• add parser-backed markdown normalization before TTS chunking
• convert common markdown constructs into readable speech-oriented plain text
• add regression tests for headings, emphasis, links, lists, code fences, and real repository markdown

Problem

Markdown selections were being sent directly to TTS, so users heard source syntax like hash hash, asterisk asterisk, and raw link/code markers instead of natural prose.

Implementation

• use pulldown-cmark for deterministic offline parsing
• normalize headings, emphasis, links, blockquotes, task lists, tables, images, and section breaks for spoken output
• simplify inline code for speech and intentionally replace fenced code blocks with Code example omitted.
• preserve punctuation and spacing needed for natural prosody

Validation

• cargo test text_normalization -- --nocapture
• cargo check

[gemini-code-assist]

Code Review

This pull request introduces a text normalization module for the TTS system, utilizing the pulldown-cmark library to convert Markdown content into speech-friendly text. Key changes include the integration of this normalization step into the TTSManager and the addition of logic to handle various Markdown elements like headings, lists, and links while omitting complex code blocks. Feedback focuses on improving the robustness of HTML entity decoding and simplifying whitespace normalization logic.

[gemini-code-assist]

This function can be simplified by using the idiomatic split_whitespace().join(" ") pattern, which is more concise and readable for collapsing and normalizing whitespace.

fn normalize_inline_whitespace(text: &str) -> String {
    text.split_whitespace().collect::<Vec<_>>().join(" ")
}

[gemini-code-assist]

The current HTML entity decoding is limited, only handling a few named entities and one specific numeric entity (#39). This can be made more robust by handling all numeric entities (both decimal and hexadecimal), which are common in HTML. This would improve the accuracy of the text normalization for a wider range of inputs.

                let decoded = if let Some(code) = entity.strip_prefix('#') {
                    let (radix, code_str) = if let Some(hex_code) = code.strip_prefix('x') {
                        (16, hex_code)
                    } else {
                        (10, code)
                    };
                    u32::from_str_radix(code_str, radix)
                        .ok()
                        .and_then(std::char::from_u32)
                        .map(|c| c.to_string())
                        .unwrap_or_default()
                } else {
                    match entity.as_str() {
                        "amp" => "&".to_string(),
                        "lt" => "<".to_string(),
                        "gt" => ">".to_string(),
                        "quot" => "\"".to_string(),
                        "apos" => "'".to_string(),
                        "nbsp" => " ".to_string(),
                        _ => String::new(),
                    }
                };
                out.push_str(&decoded);