fix(builtins): cap parallel cartesian product size to prevent memory blowup

[chaliy]

Summary

• Add MAX_CARTESIAN_PRODUCT constant (100,000) to cap combinations
• Pre-calculate total product size with checked_mul before allocating
• Return error with exit code 1 when limit exceeded
• Small cartesian products still work correctly

Test plan

• Small products (2x2=4) work correctly
• Huge products (4^20 ~1 trillion) return error
• End-to-end test via run_parallel() with 20 groups
• All existing parallel tests pass

Closes #991