perf: use the memory reserved for stack instead of allocating boxes

[joonazan]

No description provided.

[antoniolocascio]

LGTM

[github-actions]

Benchmark report

Benchmark	Symbol	Base Eff	Head Eff (%)	Base Raw	Head Raw (%)	Base Blake	Head Blake (%)	Base Bigint	Head Bigint (%)
block	run_prepared	228,452,352	228,290,736 (-0.07%)	201,173,308	201,011,692 (-0.08%)	159,100	159,100 (+0.00%)	6,183,361	6,183,361 (+0.00%)
precompiles	bn254_ecadd	55,852	55,852 (+0.00%)	50,400	50,400 (+0.00%)	0	0 (+0.00%)	1,363	1,363 (+0.00%)
precompiles	bn254_ecmul	1,040,671	1,040,671 (+0.00%)	717,727	717,727 (+0.00%)	0	0 (+0.00%)	80,736	80,736 (+0.00%)
precompiles	bn254_pairing	83,642,672	83,642,672 (+0.00%)	69,108,188	69,108,188 (+0.00%)	0	0 (+0.00%)	3,633,621	3,633,621 (+0.00%)
precompiles	ecrecover	496,651	497,600 (+0.19%)	324,955	325,824 (+0.27%)	0	0 (+0.00%)	42,924	42,944 (+0.05%)
precompiles	id	1,282	1,282 (+0.00%)	1,282	1,282 (+0.00%)	0	0 (+0.00%)	0	0 (+0.00%)
precompiles	modexp	29,127,513	29,127,513 (+0.00%)	29,127,513	29,127,513 (+0.00%)	0	0 (+0.00%)	0	0 (+0.00%)
precompiles	prepare_for_call	5,746	5,746 (+0.00%)	5,746	5,746 (+0.00%)	0	0 (+0.00%)	0	0 (+0.00%)
precompiles	process_transaction	84,722,874	84,729,295 (+0.01%)	70,018,982	70,022,715 (+0.01%)	20	20 (+0.00%)	3,675,893	3,676,565 (+0.02%)
precompiles	ripemd	8,399	8,399 (+0.00%)	8,399	8,399 (+0.00%)	0	0 (+0.00%)	0	0 (+0.00%)
precompiles	run_prepared	85,290,370	85,296,787 (+0.01%)	70,509,038	70,512,767 (+0.01%)	4,860	4,860 (+0.00%)	3,675,893	3,676,565 (+0.02%)
precompiles	sha256	13,495	13,495 (+0.00%)	13,495	13,495 (+0.00%)	0	0 (+0.00%)	0	0 (+0.00%)
precompiles	verify_and_apply_batch	223,747	223,589 (-0.07%)	189,987	189,829 (-0.08%)	2,110	2,110 (+0.00%)	0	0 (+0.00%)

[joonazan]

Actually, this probably leaks memory because the code repeatedly MaybeUninit::writes the same memory without dropping it in between.

I'm not sure what is a better design. Maybe the EEs should be in boxes, just boxes allocated in this special way.