Poor performance when using return-value optimization due to double reset, nil check, genericReset

[arnetheduck]

Notice three things:

• zeroMem happens in f2, then genericReset happens in f - these are redundant
• nil check redundant - RVO should never end up with a nil variable here
• genericReset is extremely slow, in and of itself, because it can't be analysed by C compiler

type X = object
  a, b, c: seq[int]

proc f(): X = X()

proc f2() =
  var x = f()

f2()

N_LIB_PRIVATE N_NIMCALL(void, f__KQQBzps5MRoTwkbIifdBMQ)(tyObject_X__KqwtR1mZdQ5eNccTtsN3pA* Result) {
        chckNil((void*)Result);
        genericReset((void*)Result, (&NTI__KqwtR1mZdQ5eNccTtsN3pA_));
}
N_LIB_PRIVATE N_NIMCALL(void, f2__7DOD66mksAP2O3HJgMd2CQ)(void) {
        tyObject_X__KqwtR1mZdQ5eNccTtsN3pA x;
        nimZeroMem((void*)(&x), sizeof(tyObject_X__KqwtR1mZdQ5eNccTtsN3pA));
        f__KQQBzps5MRoTwkbIifdBMQ((&x));
}

Nim Compiler Version 1.1.1 [Linux: amd64]
Compiled at 2020-03-29
Copyright (c) 2006-2020 by Andreas Rumpf

git hash: 8c719fce54544c8d14ce7c0e4639c974c1748668
active boot switches: -d:release

[mratsim]

Related: #8745