Slight code optimisations

[mkalinin]

There are a few places where slight optimisations are possible:

• precompute the record_offset = i * RECORD_SIZE, saves a couple of mul ops
• replace mul(3, head_idx) -> mul(3, i) -> add(prod1, prod2) with add(head_idx, i) -> mul(3, sum), saves a push, mul pair of ops
• remove and(0xffffffff00000000000000000000000000000000, pk[32:48]) as the next shl(12*8, pk[32:48]) does the job

[lightclient]

sorry, was kicking off the CI

[lightclient]

Nice optimizations, LGTM!