Optimize recursive calls

[Robbepop]

While trying to figure out #339 I discovered some optimization potential for recursive calls and host calls which led to a fantastic speed up.

This basically reuses the local variable function_frame inside a hot loop instead of hot swapping the most recent function frame using the call stack. This behavior was inherited by the old wasmi engine (so we could implement an improvement there as well potentially).

Benchmarks

Before

execute/tiny_keccak/v1         time: [982.21 us 985.14 us 988.71 us]
execute/rev_complement/v1      time: [1.2141 ms 1.2168 ms 1.2200 ms]
execute/regex_redux/v1         time: [1.3259 ms 1.3288 ms 1.3322 ms]
execute/count_until/v1         time: [1.7190 ms 1.7231 ms 1.7274 ms]
execute/factorial_recursive/v1 time: [1.4232 us 1.4269 us 1.4311 us]
execute/factorial_optimized/v1 time: [725.67 ns 727.49 ns 729.57 ns]
execute/recursive_ok/v1        time: [376.72 us 378.03 us 379.50 us]
execute/recursive_trap/v1      time: [39.879 us 39.987 us 40.108 us]
execute/host_calls/v1          time: [54.598 us 54.743 us 54.898 us]

After

execute/tiny_keccak/v1         time:   [988.81 us 991.87 us 995.41 us]
execute/rev_complement/v1      time:   [1.2125 ms 1.2150 ms 1.2179 ms]
execute/regex_redux/v1         time:   [1.2941 ms 1.2978 ms 1.3026 ms]
execute/count_until/v1         time:   [1.7007 ms 1.7133 ms 1.7367 ms]
execute/factorial_recursive/v1 time:   [1.2394 us 1.2433 us 1.2479 us]
execute/factorial_optimized/v1 time:   [723.59 ns 727.05 ns 732.14 ns]
execute/recursive_ok/v1        time:   [316.64 us 320.51 us 325.65 us]
execute/recursive_trap/v1      time:   [30.679 us 30.745 us 30.816 us]
execute/host_calls/v1          time:   [47.431 us 47.576 us 47.747 us]

Summary

We see improvements in

• factorial_recursive: 1.4269 us down to 1.2433 us -> 13% faster
• recursive_ok: 378.03 us down to 320.51 us -> 16% faster
• recursive_trap: 39.987 us down to 30.745 us -> 23% faster
• host_calls: 54.743 us down to 47.576 us -> 13% faster