Improvement: ZodError performance impact

[mysuf]

Removing ZodError inheritance from the Error class doubles or even triples the op/sec for invalid benchmark tests. This is because a stack trace doesn't need to be collected, which is useless for parse errors anyway. It is pretty easy-to-do modification. The only con is that it breaks compatibility e.g. result.error instanceof Error which is somehow weird so I think it is worth it.

Invalid Tests	Error Subclass (instanceof Error)	Standalone Class
z.enum: invalid	131,440 ops/sec	250,984 ops/sec
long z.enum: invalid	118,232 ops/sec	224,574 ops/sec
z.undefined: invalid	127,715 ops/sec	260,896 ops/sec
z.literal: invalid	125,889 ops/sec	252,896 ops/sec
z.number: invalid type	122,146 ops/sec	259,295 ops/sec
z.number: invalid number	125,977 ops/sec	266,057 ops/sec
z.date: invalid	124,918 ops/sec	255,595 ops/sec
z.date: invalid min	123,008 ops/sec	249,733 ops/sec
z.date: invalid max	123,179 ops/sec	248,390 ops/sec
z.symbol: invalid	143,278 ops/sec	349,873 ops/sec
z.string: invalid: null	142,573 ops/sec	344,606 ops/sec
z.object: empty: invalid: null	141,122 ops/sec	346,413 ops/sec
z.object: short: invalid: null	141,898 ops/sec	346,534 ops/sec
z.object: long: invalid: null	136,760 ops/sec	347,982 ops/sec

Another +10% op/sec (V8 optimizations) on invalids just by changing handleResult's:

{
        success: false,
        get error() {
            if (this._error)
                return this._error;
            const error = new ZodError(ctx.common.issues);
            this._error = error;
            return this._error;
        },
}

to

return {
        success: false,
        error: new ZodError(ctx.common.issues)
};

Btw I'm showing just weak spots. Not proposal code.