Compiler perf: avoid the need to share variables across the `cases` of a `match` statement

[TomWambsgans]

Currently, in order not to waste memory cells, we need to use the same variables across cases of a match statement:

match encoding[i] {
    0 => {
        var_1 = chain_tips + i;
        var_2 = public_key + i;
        var_3 = malloc_vec(1);
        var_4 = malloc_vec(1);
        poseidon16(var_1, pointer_to_zero_vector, var_3, COMPRESSION);
        poseidon16(var_3, pointer_to_zero_vector, var_4, COMPRESSION);
        poseidon16(var_4, pointer_to_zero_vector, var_2, COMPRESSION);
    }
    1 => {
        var_3 = malloc_vec(1);
        var_1 = chain_tips + i;
        var_2 = public_key + i;
        poseidon16(var_1, pointer_to_zero_vector, var_3, COMPRESSION);
        poseidon16(var_3, pointer_to_zero_vector, var_2, COMPRESSION);
    }
    2 => {
        var_1 = chain_tips + i;
        var_2 = public_key + i;
        poseidon16(var_1, pointer_to_zero_vector, var_2, COMPRESSION);
    }
    3 => {
        var_1 = chain_tips_ptr + (i * 8);
        var_2 = public_key_ptr + (i * 8);
        var_3 = var_1 + 3;
        var_4 = var_2 + 3;
        dot_product(var_1, pointer_to_one_vector * 8, var_2, 1);
        dot_product(var_3, pointer_to_one_vector * 8, var_4, 1);
    }
}

Ideally, we want the compiler to reuse the same memory cells even if variables don't have the same name.

[morganthomas]

Resolved by #103