perf: optimize make_word() to avoid unnecessary allocations

[andygrove]

Summary

Optimizes the Token::make_word() function which is called for every identifier and keyword token during tokenization.

Changes:

• Skip uppercase string allocation entirely for quoted identifiers (e.g., "myColumn", `table`) since they are never keywords
• Use to_ascii_uppercase() instead of to_uppercase() since SQL keywords are ASCII-only, avoiding Unicode case conversion overhead

Before:

let word_uppercase = word.to_uppercase();  // Always allocated
Token::Word(Word {
    value: word.to_string(),
    keyword: if quote_style.is_none() { /* lookup */ } else { Keyword::NoKeyword },
})

After:

let keyword = if quote_style.is_none() {
    let word_uppercase = word.to_ascii_uppercase();  // Only when needed
    // lookup
} else {
    Keyword::NoKeyword
};
Token::Word(Word { value: word.to_string(), keyword })

Test plan

• All existing tests pass
• cargo clippy passes

🤖 Generated with Claude Code

[iffyio]

LGTM! Thanks @andygrove!