Run parse and serialize upb benchmarks with both runtime parsed and compiled-in minitables. Some benchmarks have seen 20% swings on big cores from minor code changes that shifted the layout of minitables in relro and rodata, which is unhelpful when trying to measure smaller changes.

PiperOrigin-RevId: 816296865