Benchmark all functions receiving "special treatment" in FunC compiler against related TVM instructions

[novusnota]

For example, compare the gas usage of .load_int() in FunC or .loadInt() in Tact to an assembly function that'll use just LDIX. Do so for all functions that are marked with // special treatment in Func compiler, so not replaced with X.

Alongside those gas benchmarks, it might be nice to gather sequences of instructions executed from the Sandbox logs.

This might help with:

1. Discovering new optimizations to perform
2. Replacing FunC functions with asm functions iff we can match or outmatch them

[novusnota]

Related to #1554

[Shvandre]

I did some research on this topic and found a few interesting details:

1. The funC compiler has an optimization where, if the second argument in store is constant, it uses the STU/STI instructions instead of STX/STIX. This saves one instruction for pushing the argument onto the stack – around 18/26 gas.
2. Storing arguments in a straightforward order is often more gas-efficient than storing them across multiple lines.

For example first function is better, than the second one.

builder usualBuild() impure inline {
    builder b = begin_cell().store_uint(0, 32).store_int(1, 32);
    return b;
}

builder usualBuild2() impure inline {
    builder b = begin_cell();
    b = b.store_uint(0, 32);
    b = b.store_int(1, 32);
    return b;
}

(In this case, there are extra SWAP instructions.)

3. If we remove the STU/STI optimization and use STUX/STIX instead, we get an additional PUSH, which typically costs 26 or 34 gas, depending on the value.

In conclusion, I’d recommend not removing this optimization that funC already provides. Instead, I’d focus on optimizing these instructions further, like turning them into something like STSLICECONST #1592 , STZEROES, or STONES, which is #1561