Sets the ML_ACC boost value to 650. The medium core's capacity is 763, so this means our benchmark process will always be scheduled onto the medium or big cores. We experimented with different uclamp values, and 650 is a good balance between power and performance.

Measurements:
MLPerf IC model Latency (ms) Power (mW) Energy/inference (mJ) MLPerf Scores
Uclamp = 512 0.981 3539 3.47 826
Uclamp = 600 0.973 3683 3.58 872
Uclamp = 650 0.899 3917 3.52 970 <- good balance
Uclamp = 700 0.892 4067 3.63 976
Uclamp = 800 0.875 4440 3.89 997

https://docs.google.com/presentation/d/1zx7sLkhOClmuRTCrq8-l3N1mZrrv7f-CtcdMuzV0eaI/edit?usp=sharing

Bug: 232183574
Test: MLPerf score on P22 improved significantly.

Change-Id: Ib93ecee5af3f1cbe83c5a96dc14b3a537db5abc8