13. Allocation-Free Operators Rule Attempts and Pooled Operators Rule Frames in the Engine and VMs

• Ports the generated parsers' Operators Rule technique (change 11) to the engine and both VMs. parser.longest returns the winning operatorsRuleAttempt by value instead of allocating one per candidate, and saves captures across a reset in a reusable buffer (parser.saved) instead of a fresh slice.
• The iterative VM pools its skip, longest, nud and Operators Rule frames (generic reuse[T], released in release), holds attempts by value, and pushes no skip frame when the Operators Rule has no skip.
• Pitfall: putting the operatorsRuleAttempt into iresult (returned on every VM step) made the iterative VM about 20% slower on every grammar, Operators Rule or not. The attempt is passed through ivmMachine.attempt instead, so iresult stays small.
• Effect (min of 6, codepoints, allocs/op before → after):

  Workload	closure	bytecode	iterative
  Arith_OperatorsRule	173k → 134k	269k → 231k	484k → 231k; 79 → 67 ms
  Minilang	137k → 75k	203k → 140k	274k → 140k; 92 → 75 ms
  JSON (no Operators Rule)	unchanged	unchanged	unchanged; 107 → 101 ms