Switch Parcel to FastNative

Also fixes ParcelBenchmark to have bounded
memory usage to avoid OOM'ing during runs

Test: refactor, no behavior change

ParcelBenchmark results from bullhead
Before:
      timeReadByte 74.43ns
      timeReadInt 74.49ns
      timeReadLong 74.13ns
      timeWriteByte 81.81ns
      timeWriteInt 82.09ns
      timeWriteLong 81.84ns

After:
      timeReadByte 47.08ns
      timeReadInt 48.38ns
      timeReadLong 48.16ns
      timeWriteByte 55.90ns
      timeWriteInt 55.85ns
      timeWriteLong 56.58ns

Change-Id: I61b823d1d2beb86e00c196abd4dce65efa8fa7f0