Two changes to the hot encode loop in _ConvertBytecodeToConcrete.concrete_instructions:

1. ConcreteInstr._from_trusted factory

   ConcreteInstr(instr_name, c_arg, location=location) at the bottom of the
   loop went through the full __init__ -> _check_arg -> _set chain on every
   instruction, even though name/opcode/arg/location are all derived from
   already-validated Instr objects. A new _from_trusted classmethod bypasses
   _check_arg entirely and replicates only the size computation from _set,
   using object.__new__ + direct slot assignment — the same pattern as the
   existing _from_opcode fast path.

2. Direct private slot access

   instr.location, instr.name, and instr.arg are properties whose bodies are
   each a single "return self._xxx". Accessing instr._location, instr._name,
   and instr._arg directly avoids the property descriptor lookup on every
   iteration. This is safe because instr is a validated Instr instance at this
   point (guarded by the assert isinstance check above).

Profile data

| Hotspot | Before | After |
|---|---|---|
| concrete_instructions own | 8.07% | 5.44% |
| concrete_instructions total | 11.71% | 9.44% |
| ConcreteInstr.__init__ (child) | 1.30% total | gone |

Throughput (Bytecode.from_code().to_code() on dis module, 30 runs each)

| | Avg | Stddev |
|---|---|---|
| Before | 223.7 r/s | ±4.7 |
| After | 246.1 r/s | ±3.3 |

+22.4 r/s / +10.0% throughput improvement.