- Add dce.py: UsageAnalyzer that scans AST for used stdlib categories
- Modify stdlib.py: emit_stdlib() now accepts used_categories parameter
- Modify codegen.py: analyze programs before generating stdlib

Categories: core, object, http, fs, json, logger, string, array,
dict, regex, math, time, awk, exception, args, misc

Results:
- Simple "print(Hello)" program: 200+ lines -> 35 lines
- Complex programs: minimal change (use most categories anyway)

- Add estrlist.ct: string list operations with @awk acceleration
- Add estrlist_noawk.ct: pure bash version for comparison
- Add benchmark.sh: performance comparison script
- Add BENCHMARK.md: results showing @awk gives 93x speedup on union

Codegen fixes:
- Fix time.now() and time.ms() calls in dispatch_codegen
- Fix math.* namespace calls
- Previous fixes for class field assignment and nested constructors