feat: recognize brace expansion patterns in Word.parts

[mpecan]

Summary

• Adds post-hoc brace expansion detection ({a,b,c}, {1..10}, {a..z}) to the lexer's span tracking system
• Exposes BraceExpansion nodes in Word.parts decomposition so downstream consumers (e.g. security tools) can detect brace expansion without runtime evaluation
• S-expression output is unchanged — brace patterns remain literal in s-exp to match Parable exactly

Changes

• New module src/lexer/brace_expansion.rs — scans completed words for brace expansion patterns, skipping quoted/escaped/parameter-expansion regions via existing spans
• WordSpanKind::BraceExpansion — new span kind in word_builder.rs
• NodeKind::BraceExpansion { content } — new AST node in ast.rs
• WordSegment::BraceExpansion — new segment variant, wired through is_decomposable() (not is_sexp_relevant)
• Extracted span_to_segment() helper in sexp/word.rs — reduces build_segments length, collapses three duplicate match arms

Test plan

• 12 new unit tests in brace_expansion.rs (comma, range, mid-word, nested, empty, single element, trailing/leading comma, param expansion exclusion, adjacent, alpha range, range with step)
• 3 new decomposition tests in word_parts.rs (comma form, range form, mid-word split)
• All 168 existing tests pass (including echo {a,b,c} → (word "{a,b,c}") s-exp regression)
• cargo clippy --all-targets -- -D warnings clean

🤖 Generated with Claude Code