[Perf] Introduce MulAssign for LinearCombination

[ljedrz]

Moved from AleoNet#2460; just to recap, this PR improves the performance (both CPU and heap) of Poseidon::hash_many which is prominent in heap profiles, and it consists of 2 changes:

• introduce MulAssign<&F> for LinearCombination<F>, so that MulAssign<&Field<E>> for Field<E> can avoid going through Mul<&F> for &LinearCombination<F> (which clones the LinearCombination); a relevant benchmark is added too
• improve memory handling in Poseidon::hash_many via direct tweaks to allocations

These changes reduce the number of allocs/s in a node during a large deployment+execution by ~6% and speeds up the MulAssign<&Field<E>> for Field<E> operation by ~33%:

Field::mul_assign       time:   [43.078 ns 43.184 ns 43.301 ns]
                        change: [-33.164% -32.969% -32.787%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high mild

[ljedrz]

note: this new implementation is virtually identical to the preexisting Mul<&F> for LinearCombination that is currently being used after first going through Mul<&F> for &LinearCombination

[vicsn]

However cool the find, given everyone's concerns and the limited performance win, I suggest @howardwu to move it back to draft until we have bandwidth and appetite.

[ljedrz]

Oh I can mark it as draft it that's preferable, no problem.