Optimize Schema Package

[efectn]

A simple benchmark case such as:

func BenchmarkSimpleStructDecode(b *testing.B) {
	type S struct {
		A string  `schema:"a"`
		B int     `schema:"b"`
		C bool    `schema:"c"`
		D float64 `schema:"d"`
		E struct {
			F float64 `schema:"f"`
		} `schema:"e"`
	}
	s := S{}
	data := map[string][]string{
		"a":   {"abc"},
		"b":   {"123"},
		"c":   {"true"},
		"d":   {"3.14"},
		"e.f": {"3.14"},
	}
	decoder := NewDecoder()
	b.ResetTimer()

	for i := 0; i < b.N; i++ {
		decoder.Decode(&s, data)
	}
}

resulted with: BenchmarkSimpleStruct-16 326770 3600 ns/op 1072 B/op 58 allocs/op which is quite expensive.

      flat  flat%   sum%        cum   cum%
  110.50MB 32.69% 32.69%   137.50MB 40.68%  github.com/gofiber/schema.(*cache).parsePath
   59.50MB 17.60% 50.29%    59.50MB 17.60%  github.com/gofiber/schema.(*Decoder).findRequiredFields
   32.50MB  9.61% 59.91%    32.50MB  9.61%  reflect.(*structType).Field
   30.50MB  9.02% 68.93%    30.50MB  9.02%  reflect.New
      29MB  8.58% 77.51%    48.50MB 14.35%  github.com/gofiber/schema.(*Decoder).setDefaults
      27MB  7.99% 85.50%       27MB  7.99%  strings.genSplit
      18MB  5.33% 90.82%   337.51MB 99.85%  github.com/gofiber/schema.(*Decoder).Decode
   14.50MB  4.29% 95.11%    14.50MB  4.29%  reflect.packEface
       6MB  1.78% 96.89%        6MB  1.78%  github.com/gofiber/schema.convertString
    3.50MB  1.04% 97.92%     3.50MB  1.04%  reflect.makeFloat

these parts are the most expensive and memory-intensive parts according to pprof. I think there should be an open door to optimize parsePath, findRequiredFields and setDefaults logic. This PR is created to follow and find necessary optimizations.