improve scanner.Map's performance by caching resolved tags

[fangwentong]

scanner.Map方法通过反射获取key name，开销较大，这里通过缓存结构体的key name解析结果来提升性能。

具体表现如下，通过使用缓存可以将scanner.Map接口的吞吐提升1-2倍。

$ go test -run=BenchmarkMap -bench=BenchmarkMap -cpu=1,2,4,8 -benchtime=20000000x -benchmem
goos: darwin
goarch: amd64
pkg: github.com/didi/gendry/scanner
cpu: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
BenchmarkMapWithCache           20000000               265.4 ns/op           360 B/op          4 allocs/op
BenchmarkMapWithCache-2         20000000               152.2 ns/op           360 B/op          4 allocs/op
BenchmarkMapWithCache-4         20000000                88.22 ns/op          360 B/op          4 allocs/op
BenchmarkMapWithCache-8         20000000                70.87 ns/op          360 B/op          4 allocs/op
BenchmarkMapDisableCache        20000000               695.3 ns/op           400 B/op          9 allocs/op
BenchmarkMapDisableCache-2      20000000               348.7 ns/op           400 B/op          9 allocs/op
BenchmarkMapDisableCache-4      20000000               195.2 ns/op           400 B/op          9 allocs/op
BenchmarkMapDisableCache-8      20000000               169.8 ns/op           400 B/op          9 allocs/op
PASS
ok      github.com/didi/gendry/scanner  39.892s