Evaluate the effect of using memcpy/memmove from custom Solana SDK

[s-medvedev]

To carry out measurements, we need to use a Solana node with support for memory system calls and build a contract with a custom toolchain.

Memory operations syscall can be found in solana-labs/solana#16447

Custom toolchain supported the memory operations:

1. https://github.com/dmakarov/bpf-tools/actions/runs/748017344 - first version.
2. https://github.com/dmakarov/bpf-tools/actions/runs/755303744 - second version. This version doesn't generate a call to sol_mem* syscalls unless the number of store or load instructions is greater than 8 for the operation. It may be more efficient to perform mem-ops on small amounts of data in the VM than to make a syscall for every mem-op. It would be interesting to see your results comparing both versions of the toolchain.

Expected result:
Here is what jackmay expect:

• original toolchain should work fine but will do all the mem ops via BPF instructions within the VM
• old toolchain, should error if you program called memcmp due to bug in the toolchain
• new toolchain, should work fine, mem ops will be performed by native instructions via a syscall.

[s-medvedev]

Base	Syscalls-1	Syscalls-2
8830	7849	11%	8857	0%
9246	8199	11%	9268	0%
6293523	6230410	1%	6234533	1%
224226	95898	57%	102172	54%
9246	8199	11%	9268	0%
6074352	6054892	0%	6059066	0%
279963	236590	15%	248772	11%
9246	8199	11%	9268	0%
6018818	5999342	0%	6003465	0%
279411	236183	15%	248226	11%
9246	8199	11%	9268	0%
341660	289435	15%	294316	14%
1242029	1002169	19%	1031531	17%
107548	92136	14%	96298	10%
246454	222255	10%	233165	5%
107591	92179	14%	96341	10%
246666	222378	10%	233374	5%
300680	254991	15%	261689	13%
1743233	1602006	8%	1647768	5%
827038	748724	9%	783558	5%
107548	92136	14%	96298	10%
247376	223088	10%	234081	5%
296910	251969	15%	257596	13%
1467123	1349944	8%	1400329	5%